cont : remove /api/tags

.devops/cann.Dockerfile

@@ -5,9 +5,6 @@
 # Define the CANN base image for easier version updates later
 ARG CHIP_TYPE=910b
 ARG CANN_BASE_IMAGE=quay.io/ascend/cann:8.5.0-${CHIP_TYPE}-openeuler24.03-py3.11
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
 
 # ==============================================================================
 # BUILD STAGE
@@ -58,7 +55,6 @@ RUN mkdir -p /app/lib && \
 RUN mkdir -p /app/full && \
     cp build/bin/* /app/full/ && \
     cp *.py /app/full/ && \
-    cp -r conversion /app/full/ && \
     cp -r gguf-py /app/full/ && \
     cp -r requirements /app/full/ && \
     cp requirements.txt /app/full/
@@ -71,19 +67,6 @@ RUN mkdir -p /app/full && \
 # ==============================================================================
 FROM ${CANN_BASE_IMAGE} AS base
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
 # -- Install runtime dependencies --
 RUN yum install -y libgomp curl && \
     yum clean all && \

.devops/cpu.Dockerfile

@@ -1,7 +1,4 @@
 ARG UBUNTU_VERSION=24.04
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
 
 FROM ubuntu:$UBUNTU_VERSION AS build
 
@@ -30,7 +27,6 @@ RUN mkdir -p /app/lib && \
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
     && cp *.py /app/full \
-    && cp -r conversion /app/full \
     && cp -r gguf-py /app/full \
     && cp -r requirements /app/full \
     && cp requirements.txt /app/full \
@@ -39,19 +35,6 @@ RUN mkdir -p /app/full \
 ## Base image
 FROM ubuntu:$UBUNTU_VERSION AS base
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
 RUN apt-get update \
     && apt-get install -y libgomp1 curl \
     && apt autoremove -y \

.devops/cuda.Dockerfile

@@ -6,10 +6,6 @@ ARG BASE_CUDA_DEV_CONTAINER=nvidia/cuda:${CUDA_VERSION}-devel-ubuntu${UBUNTU_VER
 
 ARG BASE_CUDA_RUN_CONTAINER=nvidia/cuda:${CUDA_VERSION}-runtime-ubuntu${UBUNTU_VERSION}
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-
 FROM ${BASE_CUDA_DEV_CONTAINER} AS build
 
 # CUDA architecture to build for (defaults to all supported archs)
@@ -36,7 +32,6 @@ RUN mkdir -p /app/lib && \
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
     && cp *.py /app/full \
-    && cp -r conversion /app/full \
     && cp -r gguf-py /app/full \
     && cp -r requirements /app/full \
     && cp requirements.txt /app/full \
@@ -45,19 +40,6 @@ RUN mkdir -p /app/full \
 ## Base image
 FROM ${BASE_CUDA_RUN_CONTAINER} AS base
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
 RUN apt-get update \
     && apt-get install -y libgomp1 curl \
     && apt autoremove -y \

.devops/intel.Dockerfile

@@ -1,22 +1,12 @@
-ARG ONEAPI_VERSION=2025.3.3-0-devel-ubuntu24.04
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
+ARG ONEAPI_VERSION=2025.3.2-0-devel-ubuntu24.04
 
 ## Build Image
 
 FROM intel/deep-learning-essentials:$ONEAPI_VERSION AS build
 
 ARG GGML_SYCL_F16=OFF
-ARG LEVEL_ZERO_VERSION=1.28.2
-ARG LEVEL_ZERO_UBUNTU_VERSION=u24.04
 RUN apt-get update && \
-    apt-get install -y git libssl-dev wget ca-certificates && \
-    cd /tmp && \
-    wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/level-zero_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O level-zero.deb && \
-    wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/level-zero-devel_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O level-zero-devel.deb && \
-    apt-get -o Dpkg::Options::="--force-overwrite" install -y ./level-zero.deb ./level-zero-devel.deb && \
-    rm -f /tmp/level-zero.deb /tmp/level-zero-devel.deb
+    apt-get install -y git libssl-dev
 
 WORKDIR /app
 
@@ -36,7 +26,6 @@ RUN mkdir -p /app/lib && \
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
     && cp *.py /app/full \
-    && cp -r conversion /app/full \
     && cp -r gguf-py /app/full \
     && cp -r requirements /app/full \
     && cp requirements.txt /app/full \
@@ -44,24 +33,11 @@ RUN mkdir -p /app/full \
 
 FROM intel/deep-learning-essentials:$ONEAPI_VERSION AS base
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
-ARG IGC_VERSION=v2.20.5
-ARG IGC_VERSION_FULL=2_2.20.5+19972
-ARG COMPUTE_RUNTIME_VERSION=25.40.35563.10
-ARG COMPUTE_RUNTIME_VERSION_FULL=25.40.35563.10-0
-ARG IGDGMM_VERSION=22.8.2
+ARG IGC_VERSION=v2.30.1
+ARG IGC_VERSION_FULL=2_2.30.1+20950
+ARG COMPUTE_RUNTIME_VERSION=26.09.37435.1
+ARG COMPUTE_RUNTIME_VERSION_FULL=26.09.37435.1-0
+ARG IGDGMM_VERSION=22.9.0
 RUN mkdir /tmp/neo/ && cd /tmp/neo/ \
   && wget https://github.com/intel/intel-graphics-compiler/releases/download/$IGC_VERSION/intel-igc-core-${IGC_VERSION_FULL}_amd64.deb \
   && wget https://github.com/intel/intel-graphics-compiler/releases/download/$IGC_VERSION/intel-igc-opencl-${IGC_VERSION_FULL}_amd64.deb \
@@ -133,3 +109,4 @@ WORKDIR /app
 HEALTHCHECK CMD [ "curl", "-f", "http://localhost:8080/health" ]
 
 ENTRYPOINT [ "/app/llama-server" ]
+

.devops/llama-cli-cann.Dockerfile

@@ -1,7 +1,4 @@
 ARG ASCEND_VERSION=8.5.0-910b-openeuler22.03-py3.10
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
 
 FROM ascendai/cann:$ASCEND_VERSION AS build
 
@@ -31,20 +28,6 @@ RUN echo "Building with static libs" && \
 
 # TODO: use image with NNRT
 FROM ascendai/cann:$ASCEND_VERSION AS runtime
-
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
 COPY --from=build /app/build/bin/llama-cli /app/build/bin/llama-completion /
 
 ENV LC_ALL=C.utf8

.devops/musa.Dockerfile

@@ -6,10 +6,6 @@ ARG BASE_MUSA_DEV_CONTAINER=mthreads/musa:${MUSA_VERSION}-devel-ubuntu${UBUNTU_V
 
 ARG BASE_MUSA_RUN_CONTAINER=mthreads/musa:${MUSA_VERSION}-runtime-ubuntu${UBUNTU_VERSION}-amd64
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-
 FROM ${BASE_MUSA_DEV_CONTAINER} AS build
 
 # MUSA architecture to build for (defaults to all supported archs)
@@ -41,7 +37,6 @@ RUN mkdir -p /app/lib && \
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
     && cp *.py /app/full \
-    && cp -r conversion /app/full \
     && cp -r gguf-py /app/full \
     && cp -r requirements /app/full \
     && cp requirements.txt /app/full \
@@ -50,19 +45,6 @@ RUN mkdir -p /app/full \
 ## Base image
 FROM ${BASE_MUSA_RUN_CONTAINER} AS base
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
 RUN apt-get update \
     && apt-get install -y libgomp1 curl \
     && apt autoremove -y \

.devops/nix/package.nix

@@ -103,7 +103,6 @@ let
     vulkan-headers
     vulkan-loader
     shaderc
-    spirv-headers
   ];
 in
 
@@ -147,6 +146,7 @@ effectiveStdenv.mkDerivation (finalAttrs: {
       ninja
       pkg-config
       git
+      spirv-headers
     ]
     ++ optionals useCuda [
       cudaPackages.cuda_nvcc

.devops/openvino.Dockerfile

@@ -2,26 +2,10 @@ ARG OPENVINO_VERSION_MAJOR=2026.0
 ARG OPENVINO_VERSION_FULL=2026.0.0.20965.c6d6a13a886
 ARG UBUNTU_VERSION=24.04
 
-# Intel GPU driver versions. https://github.com/intel/compute-runtime/releases
-ARG IGC_VERSION=v2.30.1
-ARG IGC_VERSION_FULL=2_2.30.1+20950
-ARG COMPUTE_RUNTIME_VERSION=26.09.37435.1
-ARG COMPUTE_RUNTIME_VERSION_FULL=26.09.37435.1-0
-ARG IGDGMM_VERSION=22.9.0
-
-# Intel NPU driver versions. https://github.com/intel/linux-npu-driver/releases
-ARG NPU_DRIVER_VERSION=v1.32.0
-ARG NPU_DRIVER_FULL=v1.32.0.20260402-23905121947
-ARG LIBZE1_VERSION=1.27.0-1~24.04~ppa2
-
-# Optional proxy build arguments
+# Optional proxy build arguments - empty by default
 ARG http_proxy=
 ARG https_proxy=
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-
 ## Build Image
 FROM ubuntu:${UBUNTU_VERSION} AS build
 
@@ -81,7 +65,6 @@ RUN mkdir -p /app/lib && \
 RUN mkdir -p /app/full \
     && cp build/ReleaseOV/bin/* /app/full/ \
     && cp *.py /app/full \
-    && cp -r conversion /app/full \
     && cp -r gguf-py /app/full \
     && cp -r requirements /app/full \
     && cp requirements.txt /app/full \
@@ -93,61 +76,15 @@ FROM ubuntu:${UBUNTU_VERSION} AS base
 # Pass proxy args to runtime stage
 ARG http_proxy
 ARG https_proxy
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
 
 RUN apt-get update \
-    && apt-get install -y libgomp1 libtbb12 curl wget ocl-icd-libopencl1 \
+    && apt-get install -y libgomp1 libtbb12 curl \
     && apt autoremove -y \
     && apt clean -y \
     && rm -rf /tmp/* /var/tmp/* \
     && find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete \
     && find /var/cache -type f -delete
 
-# Install GPU drivers
-ARG IGC_VERSION
-ARG IGC_VERSION_FULL
-ARG COMPUTE_RUNTIME_VERSION
-ARG COMPUTE_RUNTIME_VERSION_FULL
-ARG IGDGMM_VERSION
-RUN mkdir /tmp/neo/ && cd /tmp/neo/ \
-    && wget https://github.com/intel/intel-graphics-compiler/releases/download/${IGC_VERSION}/intel-igc-core-${IGC_VERSION_FULL}_amd64.deb \
-    && wget https://github.com/intel/intel-graphics-compiler/releases/download/${IGC_VERSION}/intel-igc-opencl-${IGC_VERSION_FULL}_amd64.deb \
-    && wget https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/intel-ocloc-dbgsym_${COMPUTE_RUNTIME_VERSION_FULL}_amd64.ddeb \
-    && wget https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/intel-ocloc_${COMPUTE_RUNTIME_VERSION_FULL}_amd64.deb \
-    && wget https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/intel-opencl-icd-dbgsym_${COMPUTE_RUNTIME_VERSION_FULL}_amd64.ddeb \
-    && wget https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/intel-opencl-icd_${COMPUTE_RUNTIME_VERSION_FULL}_amd64.deb \
-    && wget https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/libigdgmm12_${IGDGMM_VERSION}_amd64.deb \
-    && wget https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/libze-intel-gpu1-dbgsym_${COMPUTE_RUNTIME_VERSION_FULL}_amd64.ddeb \
-    && wget https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}/libze-intel-gpu1_${COMPUTE_RUNTIME_VERSION_FULL}_amd64.deb \
-    && dpkg --install *.deb \
-    && rm -rf /tmp/neo/
-
-# Install NPU drivers
-ARG NPU_DRIVER_VERSION
-ARG NPU_DRIVER_FULL
-ARG LIBZE1_VERSION
-RUN mkdir /tmp/npu/ && cd /tmp/npu/ \
-    && wget https://github.com/intel/linux-npu-driver/releases/download/${NPU_DRIVER_VERSION}/linux-npu-driver-${NPU_DRIVER_FULL}-ubuntu2404.tar.gz \
-    && tar -xf linux-npu-driver-${NPU_DRIVER_FULL}-ubuntu2404.tar.gz \
-    && dpkg --install *.deb \
-    && rm -rf /tmp/npu/
-
-RUN cd /tmp \
-    && wget https://snapshot.ppa.launchpadcontent.net/kobuk-team/intel-graphics/ubuntu/20260324T100000Z/pool/main/l/level-zero-loader/libze1_${LIBZE1_VERSION}_amd64.deb \
-    && dpkg --install libze1_${LIBZE1_VERSION}_amd64.deb \
-    && rm libze1_${LIBZE1_VERSION}_amd64.deb
-
 COPY --from=build /app/lib/ /app/
 
 ### Full (all binaries)

.devops/rocm.Dockerfile

@@ -7,10 +7,6 @@ ARG AMDGPU_VERSION=7.2.1
 # Target the ROCm build image
 ARG BASE_ROCM_DEV_CONTAINER=rocm/dev-ubuntu-${UBUNTU_VERSION}:${ROCM_VERSION}-complete
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-
 ### Build image
 FROM ${BASE_ROCM_DEV_CONTAINER} AS build
 
@@ -53,7 +49,6 @@ RUN mkdir -p /app/lib \
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
     && cp *.py /app/full \
-    && cp -r conversion /app/full \
     && cp -r gguf-py /app/full \
     && cp -r requirements /app/full \
     && cp requirements.txt /app/full \
@@ -62,19 +57,6 @@ RUN mkdir -p /app/full \
 ## Base image
 FROM ${BASE_ROCM_DEV_CONTAINER} AS base
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
 RUN apt-get update \
     && apt-get install -y libgomp1 curl \
     && apt autoremove -y \

.devops/s390x.Dockerfile

@@ -1,8 +1,5 @@
 ARG GCC_VERSION=15.2.0
 ARG UBUNTU_VERSION=24.04
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
 
 ### Build Llama.cpp stage
 FROM gcc:${GCC_VERSION} AS build
@@ -37,7 +34,6 @@ RUN --mount=type=cache,target=/root/.ccache \
 
 COPY *.py             /opt/llama.cpp/bin
 COPY .devops/tools.sh /opt/llama.cpp/bin
-COPY conversion       /opt/llama.cpp/conversion
 
 COPY gguf-py          /opt/llama.cpp/gguf-py
 COPY requirements.txt /opt/llama.cpp/gguf-py
@@ -48,28 +44,14 @@ COPY requirements     /opt/llama.cpp/gguf-py/requirements
 FROM scratch AS collector
 
 # Copy llama.cpp binaries and libraries
-COPY --from=build /opt/llama.cpp/bin        /llama.cpp/bin
-COPY --from=build /opt/llama.cpp/lib        /llama.cpp/lib
-COPY --from=build /opt/llama.cpp/gguf-py    /llama.cpp/gguf-py
-COPY --from=build /opt/llama.cpp/conversion /llama.cpp/conversion
+COPY --from=build /opt/llama.cpp/bin     /llama.cpp/bin
+COPY --from=build /opt/llama.cpp/lib     /llama.cpp/lib
+COPY --from=build /opt/llama.cpp/gguf-py /llama.cpp/gguf-py
 
 
 ### Base image
 FROM ubuntu:${UBUNTU_VERSION} AS base
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
 RUN --mount=type=cache,target=/var/cache/apt,sharing=locked \
     --mount=type=cache,target=/var/lib/apt/lists,sharing=locked \
     apt update -y && \
@@ -109,7 +91,6 @@ RUN curl https://sh.rustup.rs -sSf | bash -s -- -y
 
 COPY --from=collector /llama.cpp/bin /app
 COPY --from=collector /llama.cpp/gguf-py /app/gguf-py
-COPY --from=collector /llama.cpp/conversion /app/conversion
 
 RUN pip install --no-cache-dir --break-system-packages \
         -r /app/gguf-py/requirements.txt

.devops/vulkan.Dockerfile

@@ -1,7 +1,4 @@
 ARG UBUNTU_VERSION=26.04
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
 
 FROM ubuntu:$UBUNTU_VERSION AS build
 
@@ -26,7 +23,6 @@ RUN mkdir -p /app/lib && \
 RUN mkdir -p /app/full \
     && cp build/bin/* /app/full \
     && cp *.py /app/full \
-    && cp -r conversion /app/full \
     && cp -r gguf-py /app/full \
     && cp -r requirements /app/full \
     && cp requirements.txt /app/full \
@@ -35,19 +31,6 @@ RUN mkdir -p /app/full \
 ## Base image
 FROM ubuntu:$UBUNTU_VERSION AS base
 
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
 RUN apt-get update \
     && apt-get install -y libgomp1 curl libvulkan1 mesa-vulkan-drivers \
     libglvnd0 libgl1 libglx0 libegl1 libgles2 \

.devops/zendnn.Dockerfile

@@ -1,101 +0,0 @@
-ARG UBUNTU_VERSION=24.04
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-
-FROM ubuntu:$UBUNTU_VERSION AS build
-
-RUN apt-get update && \
-    apt-get install -y gcc-13 g++-13 build-essential git cmake libssl-dev libomp-dev libnuma-dev python3 ca-certificates
-
-ENV CC=gcc-13 CXX=g++-13
-
-WORKDIR /app
-
-COPY . .
-
-RUN cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DLLAMA_BUILD_TESTS=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON -DGGML_ZENDNN=ON && \
-    cmake --build build -j $(nproc)
-
-RUN mkdir -p /app/lib && \
-    find build -name "*.so*" -exec cp -P {} /app/lib \;
-
-RUN mkdir -p /app/full \
-    && cp build/bin/* /app/full \
-    && cp *.py /app/full \
-    && cp -r conversion /app/full \
-    && cp -r gguf-py /app/full \
-    && cp -r requirements /app/full \
-    && cp requirements.txt /app/full \
-    && cp .devops/tools.sh /app/full/tools.sh
-
-## Base image
-FROM ubuntu:$UBUNTU_VERSION AS base
-
-ARG BUILD_DATE=N/A
-ARG APP_VERSION=N/A
-ARG APP_REVISION=N/A
-ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
-ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
-LABEL org.opencontainers.image.created=$BUILD_DATE \
-      org.opencontainers.image.version=$APP_VERSION \
-      org.opencontainers.image.revision=$APP_REVISION \
-      org.opencontainers.image.title="llama.cpp" \
-      org.opencontainers.image.description="LLM inference in C/C++" \
-      org.opencontainers.image.url=$IMAGE_URL \
-      org.opencontainers.image.source=$IMAGE_SOURCE
-
-RUN apt-get update \
-    && apt-get install -y libgomp1 libnuma1 curl \
-    && apt autoremove -y \
-    && apt clean -y \
-    && rm -rf /tmp/* /var/tmp/* \
-    && find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete \
-    && find /var/cache -type f -delete
-
-COPY --from=build /app/lib/ /app
-
-### Full
-FROM base AS full
-
-COPY --from=build /app/full /app
-
-WORKDIR /app
-
-RUN apt-get update \
-    && apt-get install -y \
-    git \
-    python3 \
-    python3-pip \
-    python3-wheel \
-    && pip install --break-system-packages --upgrade setuptools \
-    && pip install --break-system-packages -r requirements.txt \
-    && apt autoremove -y \
-    && apt clean -y \
-    && rm -rf /tmp/* /var/tmp/* \
-    && find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete \
-    && find /var/cache -type f -delete
-
-ENTRYPOINT ["/app/tools.sh"]
-
-### Light, CLI only
-FROM base AS light
-
-COPY --from=build /app/full/llama-cli /app/full/llama-completion /app
-
-WORKDIR /app
-
-ENTRYPOINT [ "/app/llama-cli" ]
-
-### Server, Server only
-FROM base AS server
-
-ENV LLAMA_ARG_HOST=0.0.0.0
-
-COPY --from=build /app/full/llama-server /app
-
-WORKDIR /app
-
-HEALTHCHECK CMD [ "curl", "-f", "http://localhost:8080/health" ]
-
-ENTRYPOINT [ "/app/llama-server" ]

.editorconfig

@@ -45,7 +45,15 @@ insert_final_newline = unset
 trim_trailing_whitespace = unset
 insert_final_newline = unset
 
-[tools/ui/**]
+[tools/server/webui/**]
+indent_style = unset
+indent_size = unset
+end_of_line = unset
+charset = unset
+trim_trailing_whitespace = unset
+insert_final_newline = unset
+
+[tools/server/public/**]
 indent_style = unset
 indent_size = unset
 end_of_line = unset

.gitattributes

@@ -0,0 +1,4 @@
+# Treat the generated single-file WebUI build as binary for diff purposes.
+# Git's pack-file delta compression still works (byte-level), but this prevents
+# git diff from printing the entire minified file on every change.
+tools/server/public/index.html -diff

.github/ISSUE_TEMPLATE/010-bug-compilation.yml

@@ -12,8 +12,6 @@ body:
         after recreating the CMake build directory and with `-DGGML_CCACHE=OFF`.
         If the compilation succeeds with ccache disabled you should be able to permanently fix the issue
         by clearing `~/.cache/ccache` (on Linux).
-
-        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
   - type: textarea
     id: commit
     attributes:

.github/ISSUE_TEMPLATE/011-bug-results.yml

@@ -1,5 +1,5 @@
 name: Bug (model use)
-description: Something goes wrong when running a model (crashes, garbled outputs, etc.).
+description: Something goes wrong when using a model (in general, not specific to a single llama.cpp module).
 title: "Eval bug: "
 labels: ["bug-unconfirmed", "model evaluation"]
 body:
@@ -12,8 +12,6 @@ body:
         If you encountered the issue while using an external UI (e.g. ollama),
         please reproduce your issue using one of the examples/binaries in this repository.
         The `llama-completion` binary can be used for simple and reproducible model inference.
-
-        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
   - type: textarea
     id: version
     attributes:
@@ -100,8 +98,8 @@ body:
       label: Relevant log output
       description: >
           Please copy and paste any relevant log output, including the command that you entered and any generated text.
-          For very long logs (thousands of lines), please upload them as files instead; the `--log-file` CLI argument can be used for this purpose.
-          On Linux you can alternatively redirect the console output of any command into a file by appending ` > llama.log 2>&1` to your command.
+          For very long logs (thousands of lines), preferably upload them as files instead.
+          On Linux you can redirect console output into a file by appending ` > llama.log 2>&1` to your command.
       value: |
         <details>
         <summary>Logs</summary>

.github/ISSUE_TEMPLATE/019-bug-misc.yml

@@ -10,8 +10,6 @@ body:
         This issue template is intended for miscellaneous bugs that don't fit into any other category.
         If you encountered the issue while using an external UI (e.g. ollama),
         please reproduce your issue using one of the examples/binaries in this repository.
-
-        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
   - type: textarea
     id: version
     attributes:
@@ -88,8 +86,8 @@ body:
       description: >
           If applicable, please copy and paste any relevant log output, including any generated text.
           If you are encountering problems specifically with the `llama_params_fit` module, always upload `--verbose` logs as well.
-          For very long logs (thousands of lines), please upload them as files instead; the `--log-file` CLI argument can be used for this purpose.
-          On Linux you can alternatively redirect the console output of any command into a file by appending ` > llama.log 2>&1` to your command.
+          For very long logs (thousands of lines), please upload them as files instead.
+          On Linux you can redirect console output into a file by appending ` > llama.log 2>&1` to your command.
       value: |
         <details>
         <summary>Logs</summary>

.github/ISSUE_TEMPLATE/020-enhancement.yml

@@ -8,8 +8,6 @@ body:
       value: |
         [Please post your idea first in Discussion if there is not yet a consensus for this enhancement request. This will help to keep this issue tracker focused on enhancements that the community has agreed needs to be implemented.](https://github.com/ggml-org/llama.cpp/discussions/categories/ideas)
 
-        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
-
   - type: checkboxes
     id: prerequisites
     attributes:

.github/ISSUE_TEMPLATE/030-research.yml

@@ -8,8 +8,6 @@ body:
       value: |
         Don't forget to check for any [duplicate research issue tickets](https://github.com/ggml-org/llama.cpp/issues?q=is%3Aopen+is%3Aissue+label%3A%22research+%F0%9F%94%AC%22)
 
-        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
-
   - type: checkboxes
     id: research-stage
     attributes:

.github/ISSUE_TEMPLATE/040-refactor.yml

@@ -9,8 +9,6 @@ body:
         Don't forget to [check for existing refactor issue tickets](https://github.com/ggml-org/llama.cpp/issues?q=is%3Aopen+is%3Aissue+label%3Arefactoring) in case it's already covered.
         Also you may want to check [Pull request refactor label as well](https://github.com/ggml-org/llama.cpp/pulls?q=is%3Aopen+is%3Apr+label%3Arefactoring) for duplicates too.
 
-        Please fill out this template yourself, copypasting language model outputs is [strictly prohibited](https://github.com/ggml-org/llama.cpp/blob/master/CONTRIBUTING.md#ai-usage-policy).
-
   - type: textarea
     id: background-description
     attributes:

.github/actions/ccache-clear/action.yml

@@ -1,22 +0,0 @@
-name: "ccache-clear"
-description: "Delete all GitHub Actions caches matching a key prefix"
-inputs:
-  key:
-    description: "Cache key prefix to match and delete"
-    required: true
-
-runs:
-  using: "composite"
-  steps:
-    - name: Clear caches
-      shell: bash
-      run: |
-        CACHES=$(gh cache list --key "ccache-${{ inputs.key }}" --json id,key --jq '.[] | "\(.id) \(.key)"' 2>/dev/null)
-        if [ -z "$CACHES" ]; then
-          echo "No caches found with key prefix: ${{ inputs.key }}"
-          exit 0
-        fi
-        while read -r id key; do
-          echo "Deleting cache: $id ($key)"
-          gh cache delete "$id"
-        done <<< "$CACHES"

.github/actions/linux-setup-spacemit/action.yml

@@ -15,6 +15,6 @@ runs:
       id: setup
       uses: ./.github/actions/unarchive-tar
       with:
-        url: https://github.com/spacemit-com/toolchain/releases/download/v${{ inputs.version }}/spacemit-toolchain-linux-glibc-x86_64-v${{ inputs.version }}.tar.xz
+        url: https://archive.spacemit.com/toolchain/spacemit-toolchain-linux-glibc-x86_64-v${{ inputs.version }}.tar.xz
         path: ${{ inputs.path }}
         strip: 1

.github/actions/unarchive-tar/action.yml

@@ -24,4 +24,4 @@ runs:
       run: |
         mkdir -p ${{ inputs.path }}
         cd ${{ inputs.path }}
-        curl --no-progress-meter -L ${{ inputs.url }} | tar -${{ inputs.type }}x --strip-components=${{ inputs.strip }}
+        curl --no-progress-meter ${{ inputs.url }} | tar -${{ inputs.type }}x --strip-components=${{ inputs.strip }}

.github/actions/windows-setup-cuda/action.yml

@@ -96,34 +96,3 @@ runs:
           echo "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
           echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
           echo "CUDA_PATH_V13_1=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.1" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
-
-    - name: Install Cuda Toolkit 13.3
-      if: ${{ inputs.cuda_version == '13.3' }}
-      shell: pwsh
-      run: |
-          mkdir -p "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3"
-          choco install unzip -y
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_crt/windows-x86_64/cuda_crt-windows-x86_64-13.3.33-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_cudart/windows-x86_64/cuda_cudart-windows-x86_64-13.3.29-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvcc/windows-x86_64/cuda_nvcc-windows-x86_64-13.3.33-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvrtc/windows-x86_64/cuda_nvrtc-windows-x86_64-13.3.33-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/libcublas/windows-x86_64/libcublas-windows-x86_64-13.5.1.27-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/libnvvm/windows-x86_64/libnvvm-windows-x86_64-13.3.33-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_nvtx/windows-x86_64/cuda_nvtx-windows-x86_64-13.3.29-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cuda_profiler_api/windows-x86_64/cuda_profiler_api-windows-x86_64-13.3.27-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/visual_studio_integration/windows-x86_64/visual_studio_integration-windows-x86_64-13.3.27-archive.zip"
-          curl -O "https://developer.download.nvidia.com/compute/cuda/redist/cccl/windows-x86_64/cccl-windows-x86_64-13.3.3.3.1-archive.zip"
-          unzip '*.zip' -d "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3"
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_crt-windows-x86_64-13.3.33-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_cudart-windows-x86_64-13.3.29-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_nvcc-windows-x86_64-13.3.33-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_nvrtc-windows-x86_64-13.3.33-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\libcublas-windows-x86_64-13.5.1.27-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\libnvvm-windows-x86_64-13.3.33-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_nvtx-windows-x86_64-13.3.29-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cuda_profiler_api-windows-x86_64-13.3.27-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\visual_studio_integration-windows-x86_64-13.3.27-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\cccl-windows-x86_64-13.3.3.3.1-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" /E /I /H /Y
-          echo "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
-          echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
-          echo "CUDA_PATH_V13_3=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8

.github/labeler.yml

@@ -73,10 +73,11 @@ android:
     - changed-files:
         - any-glob-to-any-file:
             - examples/llama.android/**
-server/ui:
+server/webui:
     - changed-files:
         - any-glob-to-any-file:
-            - tools/ui/**
+            - tools/server/webui/**
+            - tools/server/public/**
 server:
     - changed-files:
         - any-glob-to-any-file:

.github/pull_request_template.md

@@ -6,7 +6,7 @@
 
 <!-- You can provide more details and link related discussions here. Delete this section if not applicable -->
 
-## Requirements
+# Requirements
 
 <!-- IMPORTANT: Please do NOT delete this section, otherwise your PR may be rejected -->
 

.github/workflows/build-3rd-party.yml

@@ -22,9 +22,9 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
   ubuntu-24-llguidance:

.github/workflows/build-and-test-snapdragon.yml

@@ -1,148 +0,0 @@
-name: CI (snapdragon)
-
-on:
-  workflow_dispatch:
-  push:
-    branches:
-      - master
-    paths:
-      - '.github/workflows/build-and-test-snapdragon.yml'
-      - 'ggml/include/ggml-hexagon.h'
-      - 'ggml/src/ggml-hexagon/**'
-      - 'docs/backend/snapdragon/**'
-      - 'scripts/snapdragon/**'
-      - 'CMakePresets.json'
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths:
-      - '.github/workflows/build-and-test-snapdragon.yml'
-      - 'ggml/include/ggml-hexagon.h'
-      - 'ggml/src/ggml-hexagon/**'
-      - 'docs/backend/snapdragon/**'
-      - 'scripts/snapdragon/**'
-      - 'CMakePresets.json'
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-jobs:
-  android-ndk-snapdragon:
-    runs-on: ubuntu-latest
-    container:
-      image: 'ghcr.io/snapdragon-toolchain/arm64-android:v0.7'
-    defaults:
-      run:
-        shell: bash
-
-    steps:
-      - name: Clone
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-          lfs: false
-
-      - name: Build Llama.CPP for Snapdragon Android
-        id: build_llama_cpp_snapdragon_android
-        run: |
-          cp docs/backend/snapdragon/CMakeUserPresets.json .
-          cmake --preset arm64-android-snapdragon-release -B build
-          cmake --build build
-          cmake --install build --prefix pkg-snapdragon/llama.cpp
-
-      - name: Upload Llama.CPP Snapdragon Android Build Artifact
-        if: ${{ always() && steps.build_llama_cpp_snapdragon_android.outcome == 'success' }}
-        uses: actions/upload-artifact@v6
-        with:
-          name: llama-cpp-android-arm64-snapdragon
-          path: pkg-snapdragon/llama.cpp
-
-  linux-iot-snapdragon:
-    runs-on: ubuntu-latest
-    container:
-      image: 'ghcr.io/snapdragon-toolchain/arm64-linux:v0.7'
-    defaults:
-      run:
-        shell: bash
-
-    steps:
-      - name: Clone
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-          lfs: false
-
-      - name: Build Llama.CPP for Snapdragon Linux IoT
-        id: build_llama_cpp_snapdragon_linux
-        run: |
-          cp docs/backend/snapdragon/CMakeUserPresets.json .
-          cmake --preset arm64-linux-snapdragon-release -B build-snapdragon -DGGML_OPENCL=ON
-          cmake --build build-snapdragon -j $(nproc)
-          cmake --install build-snapdragon --prefix pkg-snapdragon/llama.cpp
-
-      - name: Upload Llama.CPP Snapdragon Linux IoT Build Artifact
-        if: ${{ always() && steps.build_llama_cpp_snapdragon_linux.outcome == 'success' }}
-        uses: actions/upload-artifact@v6
-        with:
-          name: llama-cpp-linux-arm64-snapdragon
-          path: pkg-snapdragon/llama.cpp
-
-  test-snapdragon-qdc:
-    name: Test on QDC Device (${{ matrix.device }})
-    needs: [android-ndk-snapdragon, linux-iot-snapdragon]
-    runs-on: ubuntu-24.04-arm
-    timeout-minutes: 90
-    strategy:
-      fail-fast: false
-      matrix:
-        device: [SM8750, SM8850, QCS9075M]
-
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v6
-
-      - name: Download build artifact
-        uses: actions/download-artifact@v7
-        with:
-          name: ${{ startsWith(matrix.device, 'QCS') && 'llama-cpp-linux-arm64-snapdragon' || 'llama-cpp-android-arm64-snapdragon' }}
-          path: pkg-snapdragon/llama.cpp
-
-      - name: Set up Python
-        uses: actions/setup-python@v6
-        with:
-          python-version: '3.x'
-          cache: pip
-
-      - name: Install system dependencies
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y curl unzip
-
-      - name: Install QDC SDK wheel
-        run: |
-          curl -fSL -o qdc_sdk.zip https://softwarecenter.qualcomm.com/api/download/software/tools/Qualcomm_Device_Cloud_SDK/All/0.2.3/qualcomm_device_cloud_sdk-0.2.3.zip
-          unzip qdc_sdk.zip -d qdc_sdk
-          pip install qdc_sdk/qualcomm_device_cloud_sdk-0.2.3-py3-none-any.whl
-
-      - name: Check QDC API key
-        id: check_secret
-        env:
-          QDC_API_KEY: ${{ secrets.QDC_API_KEY }}
-        run: echo "has-qdc-key=${{ env.QDC_API_KEY != '' }}" >> "$GITHUB_OUTPUT"
-
-      - name: Run QDC tests (${{ matrix.device }})
-        if: steps.check_secret.outputs.has-qdc-key == 'true'
-        run: |
-          python scripts/snapdragon/qdc/run_qdc_jobs.py \
-              --test       all \
-              --pkg-dir    pkg-snapdragon/llama.cpp \
-              --model-url  "https://huggingface.co/bartowski/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q4_0.gguf" \
-              --device     ${{ matrix.device }} \
-              ${{ startsWith(matrix.device, 'QCS') && '--retries 2 --retry-delay 300' || '' }}
-        env:
-          QDC_API_KEY: ${{ secrets.QDC_API_KEY }}
-
-      - name: Cleanup
-        if: always()
-        run: rm -rf pkg-snapdragon qdc_sdk qdc_sdk.zip

.github/workflows/build-android.yml

@@ -1,24 +1,26 @@
 name: CI (android)
 
 on:
-  workflow_dispatch:
+  workflow_dispatch: # allows manual triggering
   push:
     branches:
       - master
-    paths:
-      - '.github/workflows/build-android.yml'
-      - '**/CMakeLists.txt'
-      - '**/.cmake'
-      - '**/*.h'
-      - '**/*.hpp'
-      - '**/*.c'
-      - '**/*.cpp'
+    paths: [
+      '.github/workflows/build-android.yml',
+      '**/CMakeLists.txt',
+      '**/.cmake',
+      '**/*.h',
+      '**/*.hpp',
+      '**/*.c',
+      '**/*.cpp'
+    ]
 
   pull_request:
     types: [opened, synchronize, reopened]
-    paths:
-      - '.github/workflows/build-android.yml'
-      - 'examples/llama.android/**'
+    paths: [
+      '.github/workflows/build-android.yml',
+      'examples/llama.android/**'
+    ]
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
@@ -27,12 +29,12 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
-  default:
+  android:
     runs-on: ubuntu-latest
 
     steps:
@@ -58,92 +60,42 @@ jobs:
           cd examples/llama.android
           ./gradlew build --no-daemon
 
-  ndk:
+  android-ndk:
     runs-on: ubuntu-latest
     container:
       image: 'ghcr.io/snapdragon-toolchain/arm64-android:v0.3'
     defaults:
       run:
         shell: bash
-
-    steps:
-      - name: Clone
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-          lfs: false
-
-      - name: Dependencies
-        run: |
-          apt-get update
-          apt-get install -y build-essential
-
-      - name: Build
-        id: ndk_build
-        run: |
-          cmake -D ANDROID_ABI=arm64-v8a -D ANDROID_PLATFORM=android-31 -D CMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_ROOT}/build/cmake/android.toolchain.cmake -D GGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=armv8.5-a+fp16+i8mm -G Ninja -D LLAMA_OPENSSL=OFF -D GGML_OPENMP=OFF -B build
-          cmake --build build
-          cmake --install build --prefix pkg-adb/llama.cpp
-
-      - name: Upload Android Build Artifact
-        if: ${{ always() && steps.ndk_build.outcome == 'success' }}
-        uses: actions/upload-artifact@v6
-        with:
-          name: llama-cpp-android-arm64-cpu
-          path: pkg-adb/llama.cpp
-
-  arm64:
-    runs-on: ubuntu-latest
-
-    env:
-      NDK_VERSION: "29.0.14206865"
+    strategy:
+      matrix:
+        include:
+          - build: 'arm64-cpu'
+            defines: '-D ANDROID_ABI=arm64-v8a -D ANDROID_PLATFORM=android-31 -D CMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_ROOT}/build/cmake/android.toolchain.cmake -D GGML_NATIVE=OFF -DGGML_CPU_ARM_ARCH=armv8.5-a+fp16+i8mm -G Ninja -D LLAMA_OPENSSL=OFF -D GGML_OPENMP=OFF'
+          - build: 'arm64-snapdragon'
+            defines: '--preset arm64-android-snapdragon-release'
 
     steps:
       - name: Clone
         id: checkout
         uses: actions/checkout@v6
-
-      # note : disabled to spare some cache space (https://github.com/ggml-org/llama.cpp/pull/23789)
-      #        for some reason, the ccache does not improve the build time in this case
-      # example:
-      #   cache off: https://github.com/ggerganov/tmp2/actions/runs/26534713799/job/78160400831
-      #   cache on:  https://github.com/ggerganov/tmp2/actions/runs/26534713799/job/78224189394
-      #
-      #- name: ccache
-      #  uses: ggml-org/ccache-action@v1.2.21
-      #  with:
-      #    key: android-ubuntu-arm64
-      #    evict-old-files: 1d
-      #    save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Set up JDK
-        uses: actions/setup-java@v5
         with:
-          java-version: 17
-          distribution: temurin
+          fetch-depth: 0
+          lfs: false
 
-      - name: Setup Android SDK
-        uses: android-actions/setup-android@40fd30fb8d7440372e1316f5d1809ec01dcd3699 # v4.0.1
+      - name: Build Llama.CPP for Hexagon Android
+        id: build_llama_cpp_hexagon_android
+        run: |
+          if [[ "${{ matrix.build }}" == "arm64-snapdragon" ]]; then
+            cp docs/backend/snapdragon/CMakeUserPresets.json .
+          fi
+          cmake ${{ matrix.defines }} -B build
+          cmake --build build
+          cmake --install build --prefix pkg-adb/llama.cpp
+
+      - name: Upload Llama.CPP Hexagon Android Build Artifact
+        if: ${{ always() && steps.build_llama_cpp_hexagon_android.outcome == 'success' }}
+        uses: actions/upload-artifact@v6
         with:
-          log-accepted-android-sdk-licenses: false
-
-      - name: Install NDK
-        run: |
-          sdkmanager "ndk;${{ env.NDK_VERSION }}"
-          echo "ANDROID_NDK=${ANDROID_SDK_ROOT}/ndk/${{ env.NDK_VERSION }}" >> $GITHUB_ENV
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK}/build/cmake/android.toolchain.cmake \
-            -DANDROID_ABI=arm64-v8a \
-            -DANDROID_PLATFORM=android-28 \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DGGML_BACKEND_DL=ON \
-            -DGGML_NATIVE=OFF \
-            -DGGML_CPU_ALL_VARIANTS=ON \
-            -DGGML_OPENMP=OFF \
-            -DLLAMA_BUILD_BORINGSSL=ON \
-            -DGGML_RPC=ON
-          time cmake --build build --config Release -j $(nproc)
+          name: llama-cpp-android-${{ matrix.build }}
+          path: pkg-adb/llama.cpp

.github/workflows/build-apple.yml

@@ -32,12 +32,12 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
-  macos-latest-arm64:
+  macOS-latest-ios:
     runs-on: macos-latest
 
     steps:
@@ -48,80 +48,7 @@ jobs:
       - name: ccache
         uses: ggml-org/ccache-action@v1.2.21
         with:
-          key: apple-arm64
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Build
-        id: cmake_build
-        run: |
-          sysctl -a
-          cmake -B build \
-            -DCMAKE_BUILD_RPATH="@loader_path" \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_BUILD_BORINGSSL=ON \
-            -DGGML_METAL_USE_BF16=ON \
-            -DGGML_METAL_EMBED_LIBRARY=OFF \
-            -DGGML_METAL_SHADER_DEBUG=ON \
-            -DGGML_RPC=ON
-          time cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
-          leaks -atExit -- ./build/bin/test-thread-safety -hf ggml-org/gemma-3-270m-qat-GGUF -ngl 99 -p "$(printf 'hello %.0s' {1..128})" -n 16 -c 512 -ub 32 -np 2 -t 2 -lv 1
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main -E "test-llama-archs" --verbose --timeout 900
-
-  macos-latest-x64:
-    runs-on: macos-15-intel
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: apple-x64
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Build
-        id: cmake_build
-        run: |
-          sysctl -a
-          # Metal is disabled due to intermittent failures with Github runners not having a GPU:
-          # https://github.com/ggml-org/llama.cpp/actions/runs/8635935781/job/23674807267#step:5:2313
-          cmake -B build \
-            -DCMAKE_BUILD_RPATH="@loader_path" \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_BUILD_BORINGSSL=ON \
-            -DGGML_METAL=OFF \
-            -DGGML_RPC=ON \
-            -DCMAKE_OSX_DEPLOYMENT_TARGET=13.3
-          time cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main --verbose --timeout 900
-
-  macos-latest-ios:
-    runs-on: macos-latest
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      # TODO: this likely does not do anything - if yes, remove it
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: apple-ios
+          key: macOS-latest-ios
           evict-old-files: 1d
           save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
 
@@ -132,7 +59,6 @@ jobs:
           cmake -B build -G Xcode \
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
-            -DLLAMA_BUILD_APP=OFF \
             -DLLAMA_BUILD_COMMON=OFF \
             -DLLAMA_BUILD_EXAMPLES=OFF \
             -DLLAMA_BUILD_TOOLS=OFF \
@@ -163,7 +89,6 @@ jobs:
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_OPENSSL=OFF \
-            -DLLAMA_BUILD_APP=OFF \
             -DLLAMA_BUILD_EXAMPLES=OFF \
             -DLLAMA_BUILD_TOOLS=OFF \
             -DLLAMA_BUILD_TESTS=OFF \
@@ -190,7 +115,7 @@ jobs:
           xcodebuild -downloadPlatform iOS
           xcodebuild -project examples/llama.swiftui/llama.swiftui.xcodeproj -scheme llama.swiftui -sdk iphoneos CODE_SIGNING_REQUIRED=NO CODE_SIGN_IDENTITY= -destination 'generic/platform=iOS' FRAMEWORK_FOLDER_PATH=./build-ios build
 
-  macos-latest-tvos:
+  macOS-latest-tvos:
     runs-on: macos-latest
 
     steps:
@@ -198,11 +123,10 @@ jobs:
         id: checkout
         uses: actions/checkout@v6
 
-      # TODO: this likely does not do anything - if yes, remove it
       - name: ccache
         uses: ggml-org/ccache-action@v1.2.21
         with:
-          key: apple-tvos
+          key: macOS-latest-tvos
           evict-old-files: 1d
           save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
 
@@ -214,7 +138,6 @@ jobs:
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_COMMON=OFF \
-            -DLLAMA_BUILD_APP=OFF \
             -DLLAMA_BUILD_EXAMPLES=OFF \
             -DLLAMA_BUILD_TOOLS=OFF \
             -DLLAMA_BUILD_TESTS=OFF \
@@ -224,7 +147,7 @@ jobs:
             -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
           cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
 
-  macos-latest-visionos:
+  macOS-latest-visionos:
     runs-on: macos-latest
 
     steps:
@@ -232,14 +155,6 @@ jobs:
         id: checkout
         uses: actions/checkout@v6
 
-      # TODO: this likely does not do anything - if yes, remove it
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: apple-visionos
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
       - name: Build
         id: cmake_build
         run: |
@@ -248,7 +163,6 @@ jobs:
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_COMMON=OFF \
-            -DLLAMA_BUILD_APP=OFF \
             -DLLAMA_BUILD_EXAMPLES=OFF \
             -DLLAMA_BUILD_TOOLS=OFF \
             -DLLAMA_BUILD_TESTS=OFF \
@@ -258,7 +172,7 @@ jobs:
             -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
           cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
 
-  macos-latest-swift:
+  macOS-latest-swift:
     runs-on: macos-latest
     needs: macos-latest-ios-xcode
 
@@ -271,11 +185,10 @@ jobs:
         id: checkout
         uses: actions/checkout@v6
 
-      # TODO: this likely does not do anything - if yes, remove it
       - name: ccache
         uses: ggml-org/ccache-action@v1.2.21
         with:
-          key: apple-swift
+          key: macOS-latest-swift
           evict-old-files: 1d
           save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
 
@@ -293,7 +206,6 @@ jobs:
             -DGGML_METAL_USE_BF16=ON \
             -DGGML_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_OPENSSL=OFF \
-            -DLLAMA_BUILD_APP=OFF \
             -DLLAMA_BUILD_EXAMPLES=OFF \
             -DLLAMA_BUILD_TOOLS=OFF \
             -DLLAMA_BUILD_TESTS=OFF \

.github/workflows/build-cache.yml

@@ -28,7 +28,7 @@ jobs:
         id: cache-sdk
         with:
           path: ./vulkan_sdk
-          key: cache-gha-vulkan-sdk-${{ env.VULKAN_SDK_VERSION }}-${{ runner.os }}
+          key: vulkan-sdk-${{ env.VULKAN_SDK_VERSION }}-${{ runner.os }}
 
       - name: Setup Vulkan SDK
         if: steps.cache-sdk.outputs.cache-hit != 'true'
@@ -54,7 +54,7 @@ jobs:
   #      id: cache-toolchain
   #      with:
   #        path: ./spacemit_toolchain
-  #        key: cache-gha-spacemit-ime-toolchain-v${{ env.SPACEMIT_IME_TOOLCHAIN_VERSION }}-${{ runner.os }}
+  #        key: spacemit-ime-toolchain-v${{ env.SPACEMIT_IME_TOOLCHAIN_VERSION }}-${{ runner.os }}
 
   #    - name: Setup SpacemiT Toolchain
   #      if: steps.cache-toolchain.outputs.cache-hit != 'true'
@@ -81,7 +81,7 @@ jobs:
         id: cache-openvino
         with:
           path: ./openvino_toolkit
-          key: cache-gha-openvino-toolkit-v${{ env.OPENVINO_VERSION_FULL }}-${{ runner.os }}
+          key: openvino-toolkit-v${{ env.OPENVINO_VERSION_FULL }}-${{ runner.os }}
 
       - name: Setup OpenVINO Toolkit
         if: steps.cache-openvino.outputs.cache-hit != 'true'
@@ -108,7 +108,7 @@ jobs:
         id: cache-rocm
         with:
           path: C:\Program Files\AMD\ROCm
-          key: cache-gha-rocm-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ runner.os }}
+          key: rocm-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ runner.os }}
 
       - name: Setup ROCm
         if: steps.cache-rocm.outputs.cache-hit != 'true'

.github/workflows/build-cann.yml

@@ -29,76 +29,74 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
-# TODO: this build is disabled to save Github Actions resources (https://github.com/ggml-org/llama.cpp/pull/23705)
-#       in order to enable it again, we have to provision dedicated runners  to run it
-#  openEuler-latest-cann:
-#    defaults:
-#      run:
-#        shell: bash -el {0}
-#    strategy:
-#      matrix:
-#        arch: [x86, aarch64]
-#        chip_type: ['910b', '310p']
-#        build: ['Release']
-#        use_acl_graph: ['on', 'off']
-#        exclude:
-#          # 310P does not support USE_ACL_GRAPH=on
-#          - chip_type: '310p'
-#            use_acl_graph: 'on'
-#    runs-on: ${{ matrix.arch == 'aarch64' && 'ubuntu-24.04-arm' || 'ubuntu-24.04' }}
-#    steps:
-#      - name: Checkout
-#        uses: actions/checkout@v6
-#        with:
-#          fetch-depth: 0
-#
-#      - name: Free up disk space
-#        uses: ggml-org/free-disk-space@v1.3.1
-#        with:
-#          tool-cache: true
-#
-#      - name: Set container image
-#        id: cann-image
-#        run: |
-#          image="ascendai/cann:${{ matrix.chip_type == '910b' &&  '8.5.0-910b-openeuler24.03-py3.11' || '8.5.0-310p-openeuler24.03-py3.11' }}"
-#          echo "image=${image}" >> "${GITHUB_OUTPUT}"
-#
-#      - name: Pull container image
-#        run: docker pull "${{ steps.cann-image.outputs.image }}"
-#
-#      - name: Build
-#        env:
-#          BUILD_TYPE: ${{ matrix.build }}
-#          SOC_TYPE: ascend${{ matrix.chip_type }}
-#          USE_ACL_GRAPH: ${{ matrix.use_acl_graph }}
-#        run: |
-#          HOST_UID=$(id -u)
-#          HOST_GID=$(id -g)
-#
-#          docker run --rm \
-#            -v "${PWD}:/workspace" \
-#            -w /workspace \
-#            -e SOC_TYPE=${SOC_TYPE} \
-#            -e BUILD_TYPE=${BUILD_TYPE} \
-#            -e USE_ACL_GRAPH=${USE_ACL_GRAPH} \
-#            "${{ steps.cann-image.outputs.image }}" \
-#            bash -lc '
-#              set -e
-#              yum install -y --setopt=install_weak_deps=False --setopt=tsflags=nodocs git gcc gcc-c++ make cmake openssl-devel
-#              yum clean all && rm -rf /var/cache/yum
-#              git config --global --add safe.directory "/workspace"
-#              export LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${ASCEND_TOOLKIT_HOME}/$(uname -m)-linux/devlib/:${LD_LIBRARY_PATH}
-#              cmake -S . -B build \
-#                  -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
-#                  -DGGML_CANN=on \
-#                  -DSOC_TYPE=${SOC_TYPE} \
-#                  -DUSE_ACL_GRAPH=${USE_ACL_GRAPH}
-#              cmake --build build -j $(nproc)
-#
-#              chown -R '"${HOST_UID}"':'"${HOST_GID}"' /workspace/build
-#            '
+  openEuler-latest-cann:
+    defaults:
+      run:
+        shell: bash -el {0}
+    strategy:
+      matrix:
+        arch: [x86, aarch64]
+        chip_type: ['910b', '310p']
+        build: ['Release']
+        use_acl_graph: ['on', 'off']
+        exclude:
+          # 310P does not support USE_ACL_GRAPH=on
+          - chip_type: '310p'
+            use_acl_graph: 'on'
+    runs-on: ${{ matrix.arch == 'aarch64' && 'ubuntu-24.04-arm' || 'ubuntu-24.04' }}
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+
+      - name: Free up disk space
+        uses: ggml-org/free-disk-space@v1.3.1
+        with:
+          tool-cache: true
+
+      - name: Set container image
+        id: cann-image
+        run: |
+          image="ascendai/cann:${{ matrix.chip_type == '910b' &&  '8.5.0-910b-openeuler24.03-py3.11' || '8.5.0-310p-openeuler24.03-py3.11' }}"
+          echo "image=${image}" >> "${GITHUB_OUTPUT}"
+
+      - name: Pull container image
+        run: docker pull "${{ steps.cann-image.outputs.image }}"
+
+      - name: Build
+        env:
+          BUILD_TYPE: ${{ matrix.build }}
+          SOC_TYPE: ascend${{ matrix.chip_type }}
+          USE_ACL_GRAPH: ${{ matrix.use_acl_graph }}
+        run: |
+          HOST_UID=$(id -u)
+          HOST_GID=$(id -g)
+
+          docker run --rm \
+            -v "${PWD}:/workspace" \
+            -w /workspace \
+            -e SOC_TYPE=${SOC_TYPE} \
+            -e BUILD_TYPE=${BUILD_TYPE} \
+            -e USE_ACL_GRAPH=${USE_ACL_GRAPH} \
+            "${{ steps.cann-image.outputs.image }}" \
+            bash -lc '
+              set -e
+              yum install -y --setopt=install_weak_deps=False --setopt=tsflags=nodocs git gcc gcc-c++ make cmake openssl-devel
+              yum clean all && rm -rf /var/cache/yum
+              git config --global --add safe.directory "/workspace"
+              export LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${ASCEND_TOOLKIT_HOME}/$(uname -m)-linux/devlib/:${LD_LIBRARY_PATH}
+              cmake -S . -B build \
+                  -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
+                  -DGGML_CANN=on \
+                  -DSOC_TYPE=${SOC_TYPE} \
+                  -DUSE_ACL_GRAPH=${USE_ACL_GRAPH}
+              cmake --build build -j $(nproc)
+
+              chown -R '"${HOST_UID}"':'"${HOST_GID}"' /workspace/build
+            '

.github/workflows/build-cmake-pkg.yml

@@ -5,23 +5,23 @@ on:
 
 jobs:
   linux:
-    runs-on: [self-hosted, Linux, CPU]
+    runs-on: ubuntu-slim
     steps:
       - uses: actions/checkout@v6
         with:
           fetch-depth: 0
 
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y build-essential tcl cmake
+
       - name: Build
         run: |
           PREFIX="$(pwd)"/inst
-          cmake -S . -B build \
-                -DCMAKE_PREFIX_PATH="$PREFIX" \
-                -DLLAMA_OPENSSL=OFF \
-                -DLLAMA_BUILD_TESTS=OFF \
-                -DLLAMA_BUILD_TOOLS=OFF \
-                -DLLAMA_BUILD_EXAMPLES=OFF \
-                -DLLAMA_BUILD_APP=OFF \
-                -DCMAKE_BUILD_TYPE=Release
+          cmake -S . -B build -DCMAKE_PREFIX_PATH="$PREFIX" \
+                -DLLAMA_OPENSSL=OFF -DLLAMA_BUILD_TESTS=OFF -DLLAMA_BUILD_TOOLS=OFF \
+                -DLLAMA_BUILD_EXAMPLES=OFF -DCMAKE_BUILD_TYPE=Release
           cmake --build build --config Release
           cmake --install build --prefix "$PREFIX" --config Release
 

.github/workflows/build-cpu.yml

@@ -1,231 +0,0 @@
-name: CI (cpu)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-cpu.yml',
-      '.github/workflows/build-cmake-pkg.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.cu',
-      '**/*.cuh',
-      '**/*.swift',
-      '**/*.m',
-      '**/*.metal',
-      '**/*.comp',
-      '**/*.glsl',
-      '**/*.wgsl'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-cpu.yml',
-      '.github/workflows/build-cmake-pkg.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.cu',
-      '**/*.cuh',
-      '**/*.swift',
-      '**/*.m',
-      '**/*.metal',
-      '**/*.comp',
-      '**/*.glsl',
-      '**/*.wgsl'
-    ]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  build-cmake-pkg:
-    uses: ./.github/workflows/build-cmake-pkg.yml
-
-  ubuntu:
-    strategy:
-      matrix:
-        include:
-          - build: 'x64'
-            os: ubuntu-22.04
-          - build: 'arm64'
-            os: ubuntu-24.04-arm
-
-    runs-on: ${{ matrix.os }}
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: cpu-${{ matrix.os }}
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Build Dependencies
-        id: build_depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y --no-install-recommends \
-            python3 python3-pip python3-dev python3-wheel \
-            libjpeg-dev build-essential libssl-dev \
-            git-lfs
-
-      - name: Toolchain workaround (GCC 14)
-        if: ${{ contains(matrix.os, 'ubuntu-24.04') }}
-        run: |
-          sudo apt-get install -y gcc-14 g++-14
-          echo "CC=gcc-14" >> "$GITHUB_ENV"
-          echo "CXX=g++-14" >> "$GITHUB_ENV"
-
-      - name: Python Dependencies
-        id: python_depends
-        run: |
-          export PIP_BREAK_SYSTEM_PACKAGES="1"
-          python3 -m pip install --upgrade pip setuptools
-          pip3 install ./gguf-py
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DGGML_RPC=ON
-          time cmake --build build --config Release -j $(nproc)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main --verbose --timeout 900
-
-      - name: Test llama2c conversion
-        id: llama2c_test
-        run: |
-          cd build
-          echo "Fetch tokenizer"
-          wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories260K/tok512.bin
-          echo "Fetch llama2c model"
-          wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories260K/stories260K.bin
-          ./bin/llama-convert-llama2c-to-ggml --copy-vocab-from-model ./tok512.bin --llama2c-model stories260K.bin --llama2c-output-model stories260K.gguf
-          ./bin/llama-completion -m stories260K.gguf -p "One day, Lily met a Shoggoth" -n 500 -c 256
-
-  windows:
-    runs-on: windows-2025
-
-    env:
-      OPENBLAS_VERSION: 0.3.23
-      SDE_VERSION: 9.33.0-2024-01-07
-      VULKAN_VERSION: 1.4.313.2
-
-    strategy:
-      matrix:
-        include:
-          - build: 'x64-cpu-static'
-            arch: 'x64'
-            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DBUILD_SHARED_LIBS=OFF'
-          - build: 'x64-openblas'
-            arch: 'x64'
-            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON -DGGML_OPENMP=OFF -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS -DBLAS_INCLUDE_DIRS="$env:RUNNER_TEMP/openblas/include" -DBLAS_LIBRARIES="$env:RUNNER_TEMP/openblas/lib/openblas.lib"'
-          - build: 'x64-vulkan'
-            arch: 'x64'
-            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_RPC=ON -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON -DGGML_VULKAN=ON'
-          - build: 'arm64'
-            arch: 'arm64'
-            defines: '-G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON'
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: cpu-windows-2025-${{ matrix.build }}
-          variant: ccache
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Download OpenBLAS
-        id: get_openblas
-        if: ${{ matrix.build == 'x64-openblas' }}
-        run: |
-          curl.exe -o $env:RUNNER_TEMP/openblas.zip -L "https://github.com/xianyi/OpenBLAS/releases/download/v${env:OPENBLAS_VERSION}/OpenBLAS-${env:OPENBLAS_VERSION}-x64.zip"
-          curl.exe -o $env:RUNNER_TEMP/OpenBLAS.LICENSE.txt -L "https://github.com/xianyi/OpenBLAS/raw/v${env:OPENBLAS_VERSION}/LICENSE"
-          mkdir $env:RUNNER_TEMP/openblas
-          tar.exe -xvf $env:RUNNER_TEMP/openblas.zip -C $env:RUNNER_TEMP/openblas
-          $vcdir = $(vswhere -latest -products * -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property installationPath)
-          $msvc = $(join-path $vcdir $('VC\Tools\MSVC\'+$(gc -raw $(join-path $vcdir 'VC\Auxiliary\Build\Microsoft.VCToolsVersion.default.txt')).Trim()))
-          $lib =  $(join-path $msvc 'bin\Hostx64\x64\lib.exe')
-          & $lib /machine:x64 "/def:${env:RUNNER_TEMP}/openblas/lib/libopenblas.def" "/out:${env:RUNNER_TEMP}/openblas/lib/openblas.lib" /name:openblas.dll
-
-      - name: Install Vulkan SDK
-        id: get_vulkan
-        if: ${{ matrix.build == 'x64-vulkan' }}
-        run: |
-          curl.exe -o $env:RUNNER_TEMP/VulkanSDK-Installer.exe -L "https://sdk.lunarg.com/sdk/download/${env:VULKAN_VERSION}/windows/vulkansdk-windows-X64-${env:VULKAN_VERSION}.exe"
-          & "$env:RUNNER_TEMP\VulkanSDK-Installer.exe" --accept-licenses --default-answer --confirm-command install
-          Add-Content $env:GITHUB_ENV "VULKAN_SDK=C:\VulkanSDK\${env:VULKAN_VERSION}"
-          Add-Content $env:GITHUB_PATH "C:\VulkanSDK\${env:VULKAN_VERSION}\bin"
-
-      - name: Install Ninja
-        id: install_ninja
-        run: |
-          choco install ninja
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -S . -B build ${{ matrix.defines }} `
-            -DLLAMA_BUILD_BORINGSSL=ON
-          cmake --build build --config Release -j ${env:NUMBER_OF_PROCESSORS}
-
-      - name: Add libopenblas.dll
-        id: add_libopenblas_dll
-        if: ${{ matrix.build == 'x64-openblas' }}
-        run: |
-          cp $env:RUNNER_TEMP/openblas/bin/libopenblas.dll ./build/bin/Release/openblas.dll
-          cp $env:RUNNER_TEMP/OpenBLAS.LICENSE.txt ./build/bin/Release/OpenBLAS-${env:OPENBLAS_VERSION}.txt
-
-      - name: Test
-        id: cmake_test
-        if: ${{ matrix.arch == 'x64' }}
-        run: |
-          cd build
-          ctest -L main -C Release --verbose --timeout 900
-
-      # TODO: disabled for now, consider adding tests for all CPU variants instead
-      # - name: Test (Intel SDE)
-      #   id: cmake_test_sde
-      #   if: ${{ matrix.build == 'avx512-x64' && env.HAS_AVX512F == '0' }} # use Intel SDE for AVX-512 emulation
-      #   run: |
-      #     curl.exe -o $env:RUNNER_TEMP/sde.tar.xz -L "https://downloadmirror.intel.com/813591/sde-external-${env:SDE_VERSION}-win.tar.xz"
-      #     # for some weird reason windows tar doesn't like sde tar.xz
-      #     7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/sde.tar.xz
-      #     7z x "-o${env:RUNNER_TEMP}" $env:RUNNER_TEMP/sde.tar
-      #     $sde = $(join-path $env:RUNNER_TEMP sde-external-${env:SDE_VERSION}-win/sde.exe)
-      #     cd build
-      #     $env:LLAMA_SKIP_TESTS_SLOW_ON_EMULATOR = 1
-      #     & $sde -future -- ctest -L main -C Release --verbose --timeout 900

.github/workflows/build-cross.yml

@@ -277,7 +277,7 @@ jobs:
 
     env:
       # Make sure this is in sync with build-cache.yml
-      SPACEMIT_IME_TOOLCHAIN_VERSION: "1.2.4"
+      SPACEMIT_IME_TOOLCHAIN_VERSION: "1.1.2"
 
     steps:
       - uses: actions/checkout@v6
@@ -287,7 +287,7 @@ jobs:
       #  id: cache-toolchain
       #  with:
       #    path: ./spacemit_toolchain
-      #    key: cache-gha-spacemit-ime-toolchain-v${{ env.SPACEMIT_IME_TOOLCHAIN_VERSION }}-${{ runner.os }}
+      #    key: spacemit-ime-toolchain-v${{ env.SPACEMIT_IME_TOOLCHAIN_VERSION }}-${{ runner.os }}
 
       - name: Setup SpacemiT Toolchain
         #if: steps.cache-toolchain.outputs.cache-hit != 'true'
@@ -301,17 +301,16 @@ jobs:
           export RISCV_ROOT_PATH=${PWD}/spacemit_toolchain
           cmake -B build -DLLAMA_OPENSSL=OFF \
                          -DCMAKE_BUILD_TYPE=Release \
+                         -DGGML_OPENMP=OFF \
                          -DLLAMA_BUILD_EXAMPLES=ON \
-                         -DGGML_CPU_REPACK=OFF \
                          -DLLAMA_BUILD_TOOLS=ON \
                          -DLLAMA_BUILD_TESTS=OFF \
                          -DGGML_CPU_RISCV64_SPACEMIT=ON \
                          -DGGML_RVV=ON \
-                         -DGGML_RV_ZVFH=ON \
                          -DGGML_RV_ZFH=ON \
                          -DGGML_RV_ZICBOP=ON \
                          -DGGML_RV_ZIHINTPAUSE=ON \
-                         -DGGML_RV_ZBA=ON \
+                         -DRISCV64_SPACEMIT_IME_SPEC=RISCV64_SPACEMIT_IME1 \
                          -DCMAKE_TOOLCHAIN_FILE=${PWD}/cmake/riscv64-spacemit-linux-gnu-gcc.cmake
 
           cmake --build build --config Release -j $(nproc)

.github/workflows/build-cuda-ubuntu.yml

@@ -1,134 +0,0 @@
-name: CI (CUDA, ubuntu)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-cuda-ubuntu.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.cu',
-      '**/*.cuh'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-cuda-ubuntu.yml',
-      'ggml/src/ggml-cuda/**'
-    ]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  cuda:
-    runs-on: ubuntu-24.04
-    container: nvidia/cuda:12.6.2-devel-ubuntu24.04
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Install dependencies
-        env:
-          DEBIAN_FRONTEND: noninteractive
-        run: |
-          apt update
-          apt install -y cmake build-essential ninja-build libgomp1 git libssl-dev
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: cuda-ubuntu-24.04-cuda
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Build with CMake
-        # TODO: Remove GGML_CUDA_CUB_3DOT2 flag once CCCL 3.2 is bundled within CTK and that CTK version is used in this project
-        run: |
-          cmake -S . -B build -G Ninja \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DCMAKE_CUDA_ARCHITECTURES=89-real \
-            -DCMAKE_EXE_LINKER_FLAGS=-Wl,--allow-shlib-undefined \
-            -DGGML_NATIVE=OFF \
-            -DGGML_CUDA=ON \
-            -DGGML_CUDA_CUB_3DOT2=ON
-          cmake --build build
-
-  hip:
-    runs-on: ubuntu-22.04
-    container: rocm/dev-ubuntu-22.04:6.1.2
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y build-essential git cmake rocblas-dev hipblas-dev libssl-dev rocwmma-dev
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: cuda-ubuntu-22.04-hip
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Build with native CMake HIP support
-        id: cmake_build
-        run: |
-          cmake -B build -S . \
-            -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" \
-            -DGGML_HIP_ROCWMMA_FATTN=ON \
-            -DGPU_TARGETS="gfx1030" \
-            -DGGML_HIP=ON
-          cmake --build build --config Release -j $(nproc)
-
-  musa:
-    runs-on: ubuntu-22.04
-    container: mthreads/musa:rc4.3.0-devel-ubuntu22.04-amd64
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dependencies
-        id: depends
-        run: |
-          apt-get update
-          apt-get install -y build-essential git cmake libssl-dev
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: cuda-ubuntu-22.04-musa
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Build with native CMake MUSA support
-        id: cmake_build
-        run: |
-          cmake -B build -S . \
-            -DGGML_MUSA=ON
-          time cmake --build build --config Release -j $(nproc)

.github/workflows/build-cuda-windows.yml

@@ -1,162 +0,0 @@
-name: CI (CUDA, windows)
-
-# TODO: this workflow is only triggered manually because it is very heavy on the CI
-#       when we provision dedicated windows runners, we can enable it for pushes too
-# note: running this workflow manually will populate the ccache for the release builds
-#       this can be used before merging a PR to speed up the release workflow
-on:
-  workflow_dispatch: # allows manual triggering
-
-# note: this will run in queue with the release workflow
-concurrency:
-  group: release
-  queue: max
-
-env:
-  GH_TOKEN: ${{ github.token }}
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  cuda:
-    runs-on: windows-2022
-
-    permissions:
-      actions: write
-
-    strategy:
-      matrix:
-        cuda: ['12.4', '13.3']
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: release-windows-2022-x64-cuda-${{ matrix.cuda }}
-
-      - name: Install Cuda Toolkit
-        uses: ./.github/actions/windows-setup-cuda
-        with:
-          cuda_version: ${{ matrix.cuda }}
-
-      - name: Install Ninja
-        id: install_ninja
-        run: |
-          choco install ninja
-
-      - name: Build
-        id: cmake_build
-        shell: cmd
-        # TODO: Remove GGML_CUDA_CUB_3DOT2 flag once CCCL 3.2 is bundled within CTK and that CTK version is used in this project
-        run: |
-          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvarsall.bat" x64
-          cmake -S . -B build -G "Ninja Multi-Config" ^
-            -DLLAMA_BUILD_SERVER=ON ^
-            -DLLAMA_BUILD_BORINGSSL=ON ^
-            -DGGML_NATIVE=OFF ^
-            -DGGML_BACKEND_DL=ON ^
-            -DGGML_CPU_ALL_VARIANTS=ON ^
-            -DGGML_CUDA=ON ^
-            -DGGML_RPC=ON ^
-            -DGGML_CUDA_CUB_3DOT2=ON
-          set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-1
-          cmake --build build --config Release -j %NINJA_JOBS% -t ggml
-          cmake --build build --config Release
-
-      - name: ccache-clear
-        uses: ./.github/actions/ccache-clear
-        with:
-          key: release-windows-2022-x64-cuda-${{ matrix.cuda }}
-
-  hip:
-    runs-on: windows-2022
-
-    permissions:
-      actions: write
-
-    env:
-      # Make sure this is in sync with build-cache.yml
-      HIPSDK_INSTALLER_VERSION: "26.Q1"
-
-    strategy:
-      matrix:
-        include:
-          # sync with release.yml
-          - name: "radeon"
-            gpu_targets: "gfx1150;gfx1151;gfx1200;gfx1201;gfx1100;gfx1101;gfx1102;gfx1030;gfx1031;gfx1032"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Grab rocWMMA package
-        id: grab_rocwmma
-        run: |
-          curl -o rocwmma.deb "https://repo.radeon.com/rocm/apt/7.2.1/pool/main/r/rocwmma-dev/rocwmma-dev_2.2.0.70201-81~24.04_amd64.deb"
-          7z x rocwmma.deb
-          7z x data.tar
-
-      - name: Use ROCm Installation Cache
-        uses: actions/cache@v5
-        id: cache-rocm
-        with:
-          path: C:\Program Files\AMD\ROCm
-          key: cache-gha-rocm-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ runner.os }}
-
-      - name: Setup ROCm
-        if: steps.cache-rocm.outputs.cache-hit != 'true'
-        uses: ./.github/actions/windows-setup-rocm
-        with:
-          version: ${{ env.HIPSDK_INSTALLER_VERSION }}
-
-      - name: Verify ROCm
-        id: verify
-        run: |
-          # Find and test ROCm installation
-          $clangPath = Get-ChildItem 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | Select-Object -First 1
-          if (-not $clangPath) {
-            Write-Error "ROCm installation not found"
-            exit 1
-          }
-          & $clangPath.FullName --version
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          # TODO: this build does not match the build in release.yml, so we use a different cache key
-          #       ideally, the builds should match, similar to the CUDA build above so that we would be able
-          #       to populate the ccache for the release with manual runs of this workflow
-          #key: release-windows-2022-x64-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ matrix.name }}
-          key: cuda-windows-2022-x64-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ matrix.name }}
-
-      - name: Build
-        id: cmake_build
-        run: |
-          $env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path)
-          $env:CMAKE_PREFIX_PATH="${env:HIP_PATH}"
-          cmake -G "Unix Makefiles" -B build -S . `
-            -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" `
-            -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" `
-            -DCMAKE_CXX_FLAGS="-I$($PWD.Path.Replace('\', '/'))/opt/rocm-7.2.1/include/" `
-            -DCMAKE_BUILD_TYPE=Release `
-            -DLLAMA_BUILD_BORINGSSL=ON `
-            -DROCM_DIR="${env:HIP_PATH}" `
-            -DGGML_HIP=ON `
-            -DGGML_HIP_ROCWMMA_FATTN=ON `
-            -DGPU_TARGETS="gfx1100"  `
-            -DGGML_RPC=ON
-          cmake --build build -j ${env:NUMBER_OF_PROCESSORS}
-
-      - name: ccache-clear
-        uses: ./.github/actions/ccache-clear
-        with:
-          #key: release-windows-2022-x64-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ matrix.name }}
-          key: cuda-windows-2022-x64-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ matrix.name }}

.github/workflows/build-ibm.yml

@@ -1,150 +0,0 @@
-name: CI (ibm)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-ibm.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-ibm.yml',
-      'ggml/src/ggml-cpu/**'
-    ]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-
-  ubuntu-24-s390x:
-    runs-on: ubuntu-24.04-s390x
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Build Dependencies
-        id: build_depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y --no-install-recommends \
-            python3 python3-pip python3-dev python3-wheel \
-            libjpeg-dev build-essential libssl-dev \
-            git-lfs
-
-      - name: Toolchain workaround (GCC 14)
-        run: |
-          sudo apt-get install -y gcc-14 g++-14
-          echo "CC=gcc-14" >> "$GITHUB_ENV"
-          echo "CXX=g++-14" >> "$GITHUB_ENV"
-
-      - name: Python Dependencies
-        id: python_depends
-        run: |
-          export PIP_BREAK_SYSTEM_PACKAGES="1"
-          python3 -m pip install --upgrade pip setuptools
-          pip3 install ./gguf-py
-
-      - name: Swap Endianness
-        id: endianness
-        run: |
-          for f in models/*.gguf; do
-            echo YES | python3 gguf-py/gguf/scripts/gguf_convert_endian.py $f big
-          done
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DGGML_RPC=ON
-          time cmake --build build --config Release -j $(nproc)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main --verbose --timeout 900
-
-      - name: Test llama2c (s390x)
-        id: llama2c_test_s390x
-        run: |
-          cd build
-          echo "Fetch llama2c big-endian model"
-          wget https://huggingface.co/ggml-org/models/resolve/main/tinyllamas/stories260K-be.gguf
-          ./bin/llama-completion -m stories260K-be.gguf -p "One day, Lily met a Shoggoth" -n 500 -c 256
-
-  ubuntu-24-ppc64le:
-    runs-on: ubuntu-24.04-ppc64le
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Build Dependencies
-        id: build_depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y --no-install-recommends \
-            python3 python3-pip python3-dev python3-wheel \
-            libjpeg-dev build-essential libssl-dev \
-            git-lfs
-
-      - name: Toolchain workaround (GCC 14)
-        run: |
-          sudo apt-get install -y gcc-14 g++-14
-          echo "CC=gcc-14" >> "$GITHUB_ENV"
-          echo "CXX=g++-14" >> "$GITHUB_ENV"
-
-      - name: Python Dependencies
-        id: python_depends
-        run: |
-          export PIP_BREAK_SYSTEM_PACKAGES="1"
-          python3 -m pip install --upgrade pip setuptools
-          pip3 install ./gguf-py
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DGGML_RPC=ON
-          time cmake --build build --config Release -j $(nproc)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main --verbose --timeout 900
-
-      - name: Test llama2c conversion
-        id: llama2c_test
-        run: |
-          cd build
-          echo "Fetch tokenizer"
-          wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories260K/tok512.bin
-          echo "Fetch llama2c model"
-          wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories260K/stories260K.bin
-          ./bin/llama-convert-llama2c-to-ggml --copy-vocab-from-model ./tok512.bin --llama2c-model stories260K.bin --llama2c-output-model stories260K.gguf
-          ./bin/llama-completion -m stories260K.gguf -p "One day, Lily met a Shoggoth" -n 500 -c 256

.github/workflows/build-msys.yml

@@ -15,9 +15,9 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
   windows-msys2:
@@ -37,7 +37,7 @@ jobs:
       #- name: ccache
       #  uses: ggml-org/ccache-action@v1.2.16
       #  with:
-      #    key: msys-windows-2025-x64
+      #    key: windows-msys2
       #    variant: ccache
       #    evict-old-files: 1d
       #    save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}

.github/workflows/build-opencl.yml

@@ -1,82 +0,0 @@
-name: CI (opencl)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-opencl.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.cl'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-opencl.yml',
-      'ggml/src/ggml-opencl/**'
-    ]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  windows-2025-opencl-adreno:
-    runs-on: windows-2025
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: opencl-windows-2025-x64
-          variant: ccache
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Install Ninja
-        id: install_ninja
-        run: |
-          choco install ninja
-
-      - name: Install OpenCL Headers and Libs
-        id: install_opencl
-        run: |
-          git clone https://github.com/KhronosGroup/OpenCL-Headers
-          cd OpenCL-Headers
-          cmake -B build `
-            -DBUILD_TESTING=OFF `
-            -DOPENCL_HEADERS_BUILD_TESTING=OFF `
-            -DOPENCL_HEADERS_BUILD_CXX_TESTS=OFF `
-            -DCMAKE_INSTALL_PREFIX="$env:RUNNER_TEMP/opencl-arm64-release"
-          cmake --build build --target install
-          git clone https://github.com/KhronosGroup/OpenCL-ICD-Loader
-          cd OpenCL-ICD-Loader
-          cmake -B build-arm64-release `
-            -A arm64 `
-            -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/opencl-arm64-release" `
-            -DCMAKE_INSTALL_PREFIX="$env:RUNNER_TEMP/opencl-arm64-release"
-          cmake --build build-arm64-release --target install --config release
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -S . -B build -G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-windows-llvm.cmake -DCMAKE_PREFIX_PATH="$env:RUNNER_TEMP/opencl-arm64-release" -DGGML_OPENCL=ON -DGGML_OPENCL_USE_ADRENO_KERNELS=ON -DLLAMA_BUILD_BORINGSSL=ON
-          cmake --build build --config Release -j ${env:NUMBER_OF_PROCESSORS}

.github/workflows/build-openvino.yml

@@ -1,120 +0,0 @@
-name: CI (openvino)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-openvino.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-openvino.yml',
-      'ggml/src/ggml-openvino/**'
-    ]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  ubuntu-24-openvino:
-    name: ubuntu-24-openvino-${{ matrix.openvino_device }}
-
-    concurrency:
-      group: openvino-${{ matrix.variant }}-${{ github.head_ref || github.ref }}
-      cancel-in-progress: false
-
-    strategy:
-      matrix:
-        include:
-          - variant: cpu
-            runner: '"ubuntu-24.04"'
-            openvino_device: "CPU"
-          - variant: gpu
-            runner: '["self-hosted","Linux","Intel","OpenVINO"]'
-            openvino_device: "GPU"
-
-    runs-on: ${{ fromJSON(matrix.runner) }}
-
-    env:
-      # Sync versions in build-openvino.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
-      OPENVINO_VERSION_MAJOR: "2026.0"
-      OPENVINO_VERSION_FULL: "2026.0.0.20965.c6d6a13a886"
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        if: runner.environment == 'github-hosted'
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: openvino-ubuntu-24.04-${{ matrix.variant }}-no-preset-v1
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y build-essential libssl-dev libtbb12 cmake ninja-build python3-pip
-          sudo apt-get install -y ocl-icd-opencl-dev opencl-headers opencl-clhpp-headers intel-opencl-icd
-
-      - name: Use OpenVINO Toolkit Cache
-        if: runner.environment == 'github-hosted'
-        uses: actions/cache@v5
-        id: cache-openvino
-        with:
-          path: ./openvino_toolkit
-          key: cache-gha-openvino-toolkit-v${{ env.OPENVINO_VERSION_FULL }}-${{ runner.os }}
-
-      - name: Setup OpenVINO Toolkit
-        if: steps.cache-openvino.outputs.cache-hit != 'true'
-        uses: ./.github/actions/linux-setup-openvino
-        with:
-          path: ./openvino_toolkit
-          version_major: ${{ env.OPENVINO_VERSION_MAJOR }}
-          version_full: ${{ env.OPENVINO_VERSION_FULL }}
-
-      - name: Install OpenVINO dependencies
-        run: |
-          cd ./openvino_toolkit
-          chmod +x ./install_dependencies/install_openvino_dependencies.sh
-          echo "Y" | sudo -E ./install_dependencies/install_openvino_dependencies.sh
-
-      - name: Build
-        id: cmake_build
-        run: |
-          source ./openvino_toolkit/setupvars.sh
-          cmake -B build/ReleaseOV -G Ninja \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DGGML_OPENVINO=ON
-          time cmake --build build/ReleaseOV --config Release -j $(nproc)
-
-      - name: Test
-        id: cmake_test
-        # TODO: fix and re-enable the `test-llama-archs` test below
-        run: |
-          cd ${{ github.workspace }}
-          if [ "${{ matrix.openvino_device }}" = "GPU" ]; then
-            export GGML_OPENVINO_DEVICE=GPU
-          fi
-          ctest --test-dir build/ReleaseOV -L main -E "test-llama-archs" --verbose --timeout 2000

.github/workflows/build-riscv.yml

@@ -29,84 +29,11 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
-  ubuntu-cpu-riscv64-native:
-    runs-on: ubuntu-24.04-riscv
-
-    steps:
-      - name: Install dependencies
-        run: |
-          # Install necessary packages
-          sudo apt-get update
-          sudo apt-get install -y libssl-dev
-
-          # Set gcc-14 and g++-14 as the default compilers
-          sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-14 100
-          sudo update-alternatives --install /usr/bin/g++ g++ /usr/bin/g++-14 100
-
-          git lfs install
-
-      - name: Check environment
-        run: |
-          uname -a
-          gcc --version
-          g++ --version
-          ldd --version
-          cmake --version
-          rustc --version
-          env
-          echo "nproc=$(nproc)"
-
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      # note: sparing some ccache since these jobs run on dedicated runners that are not part of the organitzation
-      #- name: ccache
-      #  uses: ggml-org/ccache-action@afde29e5b5422e5da23cb1f639e8baecadeadfc3 # https://github.com/ggml-org/ccache-action/pull/1
-      #  with:
-      #    key: riscv-ubuntu-native
-      #    evict-old-files: 1d
-      #    save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DGGML_OPENMP=OFF \
-            -DLLAMA_BUILD_EXAMPLES=ON \
-            -DLLAMA_BUILD_TOOLS=ON \
-            -DLLAMA_BUILD_TESTS=ON \
-            -DCMAKE_C_COMPILER_LAUNCHER=ccache \
-            -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
-            -DGGML_RPC=ON \
-            -DCMAKE_C_COMPILER=riscv64-linux-gnu-gcc-14 \
-            -DCMAKE_CXX_COMPILER=riscv64-linux-gnu-g++-14
-
-          time cmake --build build --config Release -j $(nproc)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main --verbose --timeout 900
-
-      - name: Test llama2c conversion
-        id: llama2c_test
-        run: |
-          cd build
-          echo "Fetch tokenizer"
-          wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories260K/tok512.bin
-          echo "Fetch llama2c model"
-          wget https://huggingface.co/karpathy/tinyllamas/resolve/main/stories260K/stories260K.bin
-          ./bin/llama-convert-llama2c-to-ggml --copy-vocab-from-model ./tok512.bin --llama2c-model stories260K.bin --llama2c-output-model stories260K.gguf
-          ./bin/llama-completion -m stories260K.gguf -p "One day, Lily met a Shoggoth" -n 500 -c 256
-
   ubuntu-riscv64-native-sanitizer:
     runs-on: ubuntu-24.04-riscv
 
@@ -135,13 +62,12 @@ jobs:
         id: checkout
         uses: actions/checkout@v6
 
-      # note: sparing some ccache since these jobs run on dedicated runners that are not part of the organitzation
-      #- name: ccache
-      #  uses: ggml-org/ccache-action@afde29e5b5422e5da23cb1f639e8baecadeadfc3 # https://github.com/ggml-org/ccache-action/pull/1
-      #  with:
-      #    key: riscv-ubuntu-native-sanitizer-${{ matrix.sanitizer }}-${{ matrix.build_type }}
-      #    evict-old-files: 1d
-      #    save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
+      - name: ccache
+        uses: ggml-org/ccache-action@afde29e5b5422e5da23cb1f639e8baecadeadfc3 # https://github.com/ggml-org/ccache-action/pull/1
+        with:
+          key: ubuntu-riscv64-native-sanitizer-${{ matrix.sanitizer }}-${{ matrix.build_type }}
+          evict-old-files: 1d
+          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
 
       - name: Build
         id: cmake_build

.github/workflows/build-rpc.yml

@@ -1,66 +0,0 @@
-name: CI (rpc)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-rpc.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-rpc.yml',
-      'ggml/src/ggml-rpc/**'
-    ]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  ubuntu-latest-rpc:
-    runs-on: ubuntu-latest
-
-    continue-on-error: true
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install build-essential libssl-dev ninja-build
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -G "Ninja" \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DGGML_RPC=ON
-          time cmake --build build --config Release -j $(nproc)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main --verbose

.github/workflows/build-sanitize.yml

@@ -22,65 +22,66 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
-  ctest:
-    runs-on: [self-hosted, X64, CPU, Linux]
+  ubuntu-latest-sanitizer:
+    runs-on: ubuntu-latest
 
     continue-on-error: true
 
     strategy:
       matrix:
         sanitizer: [ADDRESS, THREAD, UNDEFINED]
+        build_type: [Debug]
 
     steps:
       - name: Clone
         id: checkout
         uses: actions/checkout@v6
 
-      # with UNDEFINED sanitizer, we have to build in Debug to avoid GCC 13 false-positive warnings
-      - name: Build (undefined)
-        id: cmake_build_undefined
-        if: ${{ matrix.sanitizer == 'UNDEFINED' }}
-        run: |
-          cmake -B build \
-            -DCMAKE_BUILD_TYPE=Debug \
-            -DLLAMA_FATAL_WARNINGS=ON \
-            -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
-            -DGGML_SANITIZE_${{ matrix.sanitizer }}=ON
+      - name: ccache
+        uses: ggml-org/ccache-action@v1.2.21
+        with:
+          key: ubuntu-latest-sanitizer-${{ matrix.sanitizer }}
+          evict-old-files: 1d
+          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
 
-          cmake --build build --config Debug -j $(nproc)
+      - name: Dependencies
+        id: depends
+        run: |
+          sudo apt-get update
+          sudo apt-get install build-essential libssl-dev
 
       - name: Build
         id: cmake_build
-        if: ${{ matrix.sanitizer == 'ADDRESS' }}
+        if: ${{ matrix.sanitizer != 'THREAD' }}
         run: |
           cmake -B build \
-            -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+            -DLLAMA_FATAL_WARNINGS=ON \
             -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
-            -DGGML_SANITIZE_${{ matrix.sanitizer }}=ON
+            -DGGML_SANITIZE_${{ matrix.sanitizer }}=ON \
+            -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
 
-          cmake --build build --config RelWithDebInfo -j $(nproc)
+          cmake --build build --config ${{ matrix.build_type }} -j $(nproc)
 
       - name: Build (no OpenMP)
         id: cmake_build_no_openmp
         if: ${{ matrix.sanitizer == 'THREAD' }}
         run: |
           cmake -B build \
-            -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+            -DLLAMA_FATAL_WARNINGS=ON \
             -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON \
             -DGGML_SANITIZE_${{ matrix.sanitizer }}=ON \
+            -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
             -DGGML_OPENMP=OFF
 
-          cmake --build build --config RelWithDebInfo -j $(nproc)
+          cmake --build build --config ${{ matrix.build_type }} -j $(nproc)
 
       - name: Test
         id: cmake_test
-        # skip run in Debug - very slow
-        if: ${{ matrix.sanitizer != 'UNDEFINED' }}
         run: |
           cd build
-          ctest -L main -E tokenizer --verbose --timeout 900
+          ctest -L main --verbose --timeout 900

.github/workflows/build-self-hosted.yml

@@ -50,12 +50,12 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
-  gpu-cuda:
+  ggml-ci-nvidia-cuda:
     runs-on: [self-hosted, Linux, NVIDIA]
 
     steps:
@@ -67,9 +67,9 @@ jobs:
         id: ggml-ci
         run: |
           nvidia-smi
-          GG_BUILD_CUDA=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+          GG_BUILD_CUDA=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
-  gpu-vulkan-nvidia-cm:
+  ggml-ci-nvidia-vulkan-cm:
     runs-on: [self-hosted, Linux, NVIDIA]
 
     steps:
@@ -81,9 +81,9 @@ jobs:
         id: ggml-ci
         run: |
           vulkaninfo --summary
-          GG_BUILD_VULKAN=1 GGML_VK_DISABLE_COOPMAT2=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+          GG_BUILD_VULKAN=1 GGML_VK_DISABLE_COOPMAT2=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
-  gpu-vulkan-nvidia-cm2:
+  ggml-ci-nvidia-vulkan-cm2:
     runs-on: [self-hosted, Linux, NVIDIA, COOPMAT2]
 
     steps:
@@ -95,39 +95,40 @@ jobs:
         id: ggml-ci
         run: |
           vulkaninfo --summary
-          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+          GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
-  gpu-webgpu-nvidia:
-    runs-on: [self-hosted, Linux, NVIDIA, X64]
+  # TODO: investigate slight precision issues in some operations for test-backend-ops on the WebGPU backend.
+  #ggml-ci-nvidia-webgpu:
+  #  runs-on: [self-hosted, Linux, NVIDIA]
 
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
+  #  steps:
+  #    - name: Clone
+  #      id: checkout
+  #      uses: actions/checkout@v6
 
-      - name: Dawn Dependency
-        id: dawn-depends
-        run: |
-          DAWN_VERSION="v20260317.182325"
-          DAWN_OWNER="google"
-          DAWN_REPO="dawn"
-          DAWN_ASSET_NAME="Dawn-18eb229ef5f707c1464cc581252e7603c73a3ef0-ubuntu-latest-Release"
-          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          curl -L -o artifact.tar.gz \
-            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          mkdir dawn
-          tar -xvf artifact.tar.gz -C dawn --strip-components=1
+  #    - name: Dawn Dependency
+  #      id: dawn-depends
+  #      run: |
+  #        DAWN_VERSION="v20260317.182325"
+  #        DAWN_OWNER="google"
+  #        DAWN_REPO="dawn"
+  #        DAWN_ASSET_NAME="Dawn-18eb229ef5f707c1464cc581252e7603c73a3ef0-ubuntu-latest-Release"
+  #        echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+  #        curl -L -o artifact.tar.gz \
+  #          "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
+  #        mkdir dawn
+  #        tar -xvf artifact.tar.gz -C dawn --strip-components=1
 
-      - name: Test
-        id: ggml-ci
-        run: |
-          GG_BUILD_WEBGPU=1 \
-          GG_BUILD_WEBGPU_DAWN_PREFIX="$GITHUB_WORKSPACE/dawn" \
-          GG_BUILD_WEBGPU_DAWN_DIR="$GITHUB_WORKSPACE/dawn/lib64/cmake/Dawn" \
-            bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+  #    - name: Test
+  #      id: ggml-ci
+  #      run: |
+  #        GG_BUILD_WEBGPU=1 \
+  #        GG_BUILD_WEBGPU_DAWN_PREFIX="$GITHUB_WORKSPACE/dawn" \
+  #        GG_BUILD_WEBGPU_DAWN_DIR="$GITHUB_WORKSPACE/dawn/lib64/cmake/Dawn" \
+  #          bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
   # TODO: provision AMX-compatible machine
-  #cpu-amx:
+  #ggml-ci-cpu-amx:
   #  runs-on: [self-hosted, Linux, CPU, AMX]
 
   #  steps:
@@ -138,10 +139,10 @@ jobs:
   #    - name: Test
   #      id: ggml-ci
   #      run: |
-  #        bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+  #        bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
   # TODO: provision AMD GPU machine
-  # amd-vulkan:
+  # ggml-ci-amd-vulkan:
   #   runs-on: [self-hosted, Linux, AMD]
 
   #   steps:
@@ -153,10 +154,10 @@ jobs:
   #       id: ggml-ci
   #       run: |
   #         vulkaninfo --summary
-  #         GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+  #         GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
   # TODO: provision AMD GPU machine
-  # amd-rocm:
+  # ggml-ci-amd-rocm:
   #   runs-on: [self-hosted, Linux, AMD]
 
   #   steps:
@@ -168,9 +169,9 @@ jobs:
   #       id: ggml-ci
   #       run: |
   #         amd-smi static
-  #         GG_BUILD_ROCM=1 GG_BUILD_AMDGPU_TARGETS="gfx1101" bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
+  #         GG_BUILD_ROCM=1 GG_BUILD_AMDGPU_TARGETS="gfx1101" bash ./ci/run.sh ~/results/llama.cpp /mnt/llama.cpp
 
-  gpu-metal:
+  ggml-ci-mac-metal:
     runs-on: [self-hosted, macOS, ARM64]
 
     steps:
@@ -183,7 +184,7 @@ jobs:
         run: |
           GG_BUILD_METAL=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
 
-  gpu-webgpu-apple:
+  ggml-ci-mac-webgpu:
     runs-on: [self-hosted, macOS, ARM64]
 
     steps:
@@ -210,7 +211,7 @@ jobs:
           GG_BUILD_WEBGPU=1 GG_BUILD_WEBGPU_DAWN_PREFIX="$GITHUB_WORKSPACE/dawn" \
             bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
 
-  gpu-vulkan:
+  ggml-ci-mac-vulkan:
     runs-on: [self-hosted, macOS, ARM64]
 
     steps:
@@ -224,7 +225,7 @@ jobs:
           vulkaninfo --summary
           GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
 
-  gpu-vulkan-intel-linux:
+  ggml-ci-linux-intel-vulkan:
     runs-on: [self-hosted, Linux, Intel]
 
     steps:
@@ -240,7 +241,7 @@ jobs:
           vulkaninfo --summary
           GG_BUILD_VULKAN=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
 
-  gpu-vulkan-intel-windows:
+  ggml-ci-win-intel-vulkan:
     runs-on: [self-hosted, Windows, X64, Intel]
 
     steps:
@@ -261,13 +262,9 @@ jobs:
           # a valid python environment for testing
           LLAMA_FATAL_WARNINGS=OFF GG_BUILD_NINJA=1 GG_BUILD_VULKAN=1 GG_BUILD_LOW_PERF=1 ./ci/run.sh ./results/llama.cpp ./mnt/llama.cpp
 
-  cpu-openvino-low-perf:
+  ggml-ci-intel-openvino-gpu-low-perf:
     runs-on: [self-hosted, Linux, Intel, OpenVINO]
 
-    concurrency:
-      group: openvino-gpu-${{ github.head_ref || github.ref }}
-      cancel-in-progress: false
-
     env:
       # Sync versions in build.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
       OPENVINO_VERSION_MAJOR: "2026.0"
@@ -295,110 +292,4 @@ jobs:
         id: ggml-ci
         run: |
           source ./openvino_toolkit/setupvars.sh
-          GG_BUILD_OPENVINO=1 GGML_OPENVINO_DEVICE=GPU GG_BUILD_LOW_PERF=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
-
-  cpu-any-low-perf:
-    runs-on: [self-hosted, CPU]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          LLAMA_ARG_THREADS=$(nproc) GG_BUILD_LOW_PERF=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
-
-  cpu-any-high-perf:
-    runs-on: [self-hosted, CPU]
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          LLAMA_ARG_THREADS=$(nproc) GG_BUILD_HIGH_PERF=1 GG_BUILD_NO_SVE=1 GG_BUILD_NO_BF16=1 GG_BUILD_EXTRA_TESTS_0=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
-
-  cpu-arm64-graviton4:
-    runs-on: ah-ubuntu_22_04-c8g_8x
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dependencies
-        id: depends
-        run: |
-          set -euxo pipefail
-          sudo apt-get update
-          sudo DEBIAN_FRONTEND=noninteractive NEEDRESTART_MODE=a \
-          apt-get install -y \
-          build-essential \
-          python3-venv \
-          gpg \
-          wget \
-          time \
-          git-lfs
-
-          git lfs install
-
-          # install the latest cmake
-          sudo install -d /usr/share/keyrings
-          wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc \
-            | gpg --dearmor \
-            | sudo tee /usr/share/keyrings/kitware-archive-keyring.gpg >/dev/null
-          echo 'deb [signed-by=/usr/share/keyrings/kitware-archive-keyring.gpg] https://apt.kitware.com/ubuntu/ jammy main' \
-            | sudo tee /etc/apt/sources.list.d/kitware.list
-          sudo apt-get update
-          sudo apt-get install -y cmake
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          LLAMA_ARG_THREADS=$(nproc) GG_BUILD_NO_BF16=1 GG_BUILD_EXTRA_TESTS_0=1 bash ./ci/run.sh ~/results/llama.cpp ~/mnt/llama.cpp
-
-  cpu-arm64-graviton4-kleidiai:
-    runs-on: ah-ubuntu_22_04-c8g_8x
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dependencies
-        id: depends
-        run: |
-          set -euxo pipefail
-          sudo apt-get update
-          sudo DEBIAN_FRONTEND=noninteractive NEEDRESTART_MODE=a \
-          apt-get install -y \
-          build-essential \
-          python3-venv \
-          gpg \
-          wget \
-          time \
-          git-lfs
-
-          git lfs install
-
-          # install the latest cmake
-          sudo install -d /usr/share/keyrings
-          wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc \
-            | gpg --dearmor \
-            | sudo tee /usr/share/keyrings/kitware-archive-keyring.gpg >/dev/null
-          echo 'deb [signed-by=/usr/share/keyrings/kitware-archive-keyring.gpg] https://apt.kitware.com/ubuntu/ jammy main' \
-            | sudo tee /etc/apt/sources.list.d/kitware.list
-          sudo apt-get update
-          sudo apt-get install -y cmake
-
-      - name: Test
-        id: ggml-ci
-        run: |
-          GG_BUILD_KLEIDIAI=1 \
-          GG_BUILD_EXTRA_TESTS_0=1 \
-          bash ./ci/run.sh ./tmp/results ./tmp/mnt
+          GG_BUILD_OPENVINO=1 GGML_OPENVINO_DEVICE=GPU GG_BUILD_LOW_PERF=1 bash ./ci/run.sh ./tmp/results ./tmp/mnt

.github/workflows/build-sycl.yml

@@ -1,162 +0,0 @@
-name: CI (sycl)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-sycl.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-sycl.yml',
-      'ggml/src/ggml-sycl/**'
-    ]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-
-# TODO: this build is disabled to save Github Actions resources (https://github.com/ggml-org/llama.cpp/pull/23705)
-#       in order to enable it again, we have to provision dedicated runners  to run it
-#  ubuntu-24-sycl:
-#    strategy:
-#      matrix:
-#        build: [fp32]
-#        include:
-#          - build: fp32
-#            fp16: OFF
-#
-#    runs-on: ubuntu-24.04
-#
-#    env:
-#      ONEAPI_ROOT: /opt/intel/oneapi/
-#      ONEAPI_INSTALLER_VERSION: "2025.3.3"
-#      LEVEL_ZERO_VERSION: "1.28.2"
-#      LEVEL_ZERO_UBUNTU_VERSION: "u24.04"
-#
-#    continue-on-error: true
-#
-#    steps:
-#      - uses: actions/checkout@v6
-#
-#      - name: Use oneAPI Installation Cache
-#        uses: actions/cache@v5
-#        id: cache-sycl
-#        with:
-#          path: ${{ env.ONEAPI_ROOT }}
-#          key: cache-gha-oneAPI-${{ env.ONEAPI_INSTALLER_VERSION }}-${{ runner.os }}
-#
-#      - name: Download & Install oneAPI
-#        shell: bash
-#        if: steps.cache-sycl.outputs.cache-hit != 'true'
-#        run: |
-#          cd /tmp
-#          wget https://registrationcenter-download.intel.com/akdlm/IRC_NAS/56f7923a-adb8-43f3-8b02-2b60fcac8cab/intel-deep-learning-essentials-2025.3.3.16_offline.sh -O intel-deep-learning-essentials_offline.sh
-#          sudo bash intel-deep-learning-essentials_offline.sh -s -a --silent --eula accept
-#
-#      - name: Install Level Zero SDK
-#        shell: bash
-#        run: |
-#          cd /tmp
-#          wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/level-zero_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O level-zero.deb
-#          wget -q "https://github.com/oneapi-src/level-zero/releases/download/v${LEVEL_ZERO_VERSION}/level-zero-devel_${LEVEL_ZERO_VERSION}%2B${LEVEL_ZERO_UBUNTU_VERSION}_amd64.deb" -O level-zero-devel.deb
-#          sudo apt-get install -y ./level-zero.deb ./level-zero-devel.deb
-#
-#      - name: Clone
-#        id: checkout
-#        uses: actions/checkout@v6
-#
-#      - name: ccache
-#        uses: ggml-org/ccache-action@v1.2.21
-#        with:
-#          key: sycl-ubuntu-24-${{ matrix.build }}
-#          evict-old-files: 1d
-#          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-#
-#      - name: Build
-#        id: cmake_build
-#        run: |
-#          source /opt/intel/oneapi/setvars.sh
-#          cmake -B build \
-#            -G "Ninja" \
-#            -DCMAKE_BUILD_TYPE=Release \
-#            -DGGML_SYCL=ON \
-#            -DCMAKE_C_COMPILER=icx \
-#            -DCMAKE_CXX_COMPILER=icpx \
-#            -DLLAMA_OPENSSL=OFF \
-#            -DGGML_NATIVE=OFF \
-#            -DGGML_SYCL_F16=${{ matrix.fp16 }}
-#          time cmake --build build --config Release -j $(nproc)
-
-# TODO: this build is disabled to save Github Actions resources (https://github.com/ggml-org/llama.cpp/pull/23705)
-#       in order to enable it again, we have to provision dedicated runners  to run it
-#  windows-latest-sycl:
-#    runs-on: windows-2022
-#
-#    defaults:
-#      run:
-#        shell: bash
-#
-#    env:
-#      WINDOWS_BASEKIT_URL: https://registrationcenter-download.intel.com/akdlm/IRC_NAS/b60765d1-2b85-4e85-86b6-cb0e9563a699/intel-deep-learning-essentials-2025.3.3.18_offline.exe
-#      WINDOWS_DPCPP_MKL: intel.oneapi.win.cpp-dpcpp-common:intel.oneapi.win.mkl.devel:intel.oneapi.win.dnnl:intel.oneapi.win.tbb.devel
-#      LEVEL_ZERO_SDK_URL: https://github.com/oneapi-src/level-zero/releases/download/v1.28.2/level-zero-win-sdk-1.28.2.zip
-#      ONEAPI_ROOT: "C:/Program Files (x86)/Intel/oneAPI"
-#      ONEAPI_INSTALLER_VERSION: "2025.3.3"
-#    steps:
-#      - name: Clone
-#        id: checkout
-#        uses: actions/checkout@v6
-#
-#      - name: Use oneAPI Installation Cache
-#        uses: actions/cache@v5
-#        id: cache-sycl
-#        with:
-#          path: ${{ env.ONEAPI_ROOT }}
-#          key: cache-gha-oneAPI-${{ env.ONEAPI_INSTALLER_VERSION }}-${{ runner.os }}
-#
-#      - name: Download & Install oneAPI
-#        shell: bash
-#        if: steps.cache-sycl.outputs.cache-hit != 'true'
-#        run: |
-#          scripts/install-oneapi.bat $WINDOWS_BASEKIT_URL $WINDOWS_DPCPP_MKL
-#
-#      - name: Install Level Zero SDK
-#        shell: pwsh
-#        run: |
-#          Invoke-WebRequest -Uri "${{ env.LEVEL_ZERO_SDK_URL }}" -OutFile "level-zero-win-sdk.zip"
-#          Expand-Archive -Path "level-zero-win-sdk.zip" -DestinationPath "C:/level-zero-sdk" -Force
-#          "LEVEL_ZERO_V1_SDK_PATH=C:/level-zero-sdk" | Out-File -FilePath $env:GITHUB_ENV -Append
-#
-#      - name: ccache
-#        uses: ggml-org/ccache-action@v1.2.21
-#        with:
-#          key: sycl-windows-latest
-#          variant: ccache
-#          evict-old-files: 1d
-#          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-#
-#      # TODO: add ssl support ; we will also need to modify win-build-sycl.bat to accept user-specified args
-#
-#      - name: Build
-#        id: cmake_build
-#        run:  examples/sycl/win-build-sycl.bat

.github/workflows/build-virtgpu.yml

@@ -1,50 +0,0 @@
-name: CI (virtgpu)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-virtgpu.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-virtgpu.yml',
-      'ggml/src/ggml-virtgpu/**'
-    ]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-jobs:
-  ubuntu-24-virtgpu:
-    runs-on: ${{ 'ubuntu-24.04-arm' || 'ubuntu-24.04' }}
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y build-essential libdrm-dev pkg-config libssl-dev
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build \
-            -DGGML_VIRTGPU=ON \
-            -DGGML_VIRTGPU_BACKEND=ON
-          cmake --build build --config Release -j $(nproc)

.github/workflows/build-vulkan.yml

@@ -31,57 +31,12 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
-  ubuntu:
-    strategy:
-      matrix:
-        include:
-          - build: 'x64'
-            os: ubuntu-24.04
-          - build: 'arm64'
-            os: ubuntu-24.04-arm
-
-    runs-on: ${{ matrix.os }}
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y gcc-14 g++-14 build-essential glslc libvulkan-dev spirv-headers libssl-dev ninja-build
-          echo "CC=gcc-14" >> "$GITHUB_ENV"
-          echo "CXX=g++-14" >> "$GITHUB_ENV"
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: vulkan-${{ matrix.os }}-new
-          variant: ccache
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Configure
-        id: cmake_configure
-        run: |
-          cmake -B build \
-            -G "Ninja" \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DGGML_VULKAN=ON
-
-      - name: Build
-        id: cmake_build
-        run: |
-          time cmake --build build -j $(nproc)
-
-  ubuntu-llvmpipe:
+  ubuntu-24-vulkan-llvmpipe:
     runs-on: ubuntu-24.04
 
     steps:
@@ -89,6 +44,13 @@ jobs:
         id: checkout
         uses: actions/checkout@v6
 
+      - name: ccache
+        uses: ggml-org/ccache-action@v1.2.21
+        with:
+          key: ubuntu-24-vulkan-llvmpipe
+          evict-old-files: 1d
+          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
+
       - name: Dependencies
         id: depends
         run: |
@@ -106,7 +68,7 @@ jobs:
         id: cache-sdk
         with:
           path: ./vulkan_sdk
-          key: cache-gha-vulkan-sdk-${{ env.VULKAN_SDK_VERSION }}-${{ runner.os }}
+          key: vulkan-sdk-${{ env.VULKAN_SDK_VERSION }}-${{ runner.os }}
 
       - name: Setup Vulkan SDK
         if: steps.cache-sdk.outputs.cache-hit != 'true'
@@ -115,13 +77,6 @@ jobs:
           path: ./vulkan_sdk
           version: ${{ env.VULKAN_SDK_VERSION }}
 
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: vulkan-ubuntu-24.04-llvmpipe
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
       - name: Build
         id: cmake_build
         run: |

.github/workflows/build-webgpu.yml

@@ -1,181 +0,0 @@
-name: CI (webgpu)
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/build-webgpu.yml',
-      '**/CMakeLists.txt',
-      '**/.cmake',
-      '**/*.h',
-      '**/*.hpp',
-      '**/*.c',
-      '**/*.cpp',
-      '**/*.wgsl'
-    ]
-
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/build-webgpu.yml',
-      'ggml/src/ggml-webgpu/**'
-    ]
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-env:
-  GGML_NLOOP: 3
-  GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-
-jobs:
-  macos:
-    runs-on: macos-latest
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: webgpu-macos-latest
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Dawn Dependency
-        id: dawn-depends
-        run: |
-          DAWN_VERSION="v20260317.182325"
-          DAWN_OWNER="google"
-          DAWN_REPO="dawn"
-          DAWN_ASSET_NAME="Dawn-18eb229ef5f707c1464cc581252e7603c73a3ef0-macos-latest-Release"
-          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          curl -L -o artifact.tar.gz \
-            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          mkdir dawn
-          tar -xvf artifact.tar.gz -C dawn --strip-components=1
-
-      - name: Build
-        id: cmake_build
-        run: |
-          export CMAKE_PREFIX_PATH=dawn
-          cmake -B build -G "Ninja" -DCMAKE_BUILD_TYPE=Release -DGGML_WEBGPU=ON -DGGML_METAL=OFF -DGGML_BLAS=OFF
-          time cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          ctest -L main --verbose --timeout 900
-
-  ubuntu:
-    runs-on: ubuntu-24.04
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: webgpu-ubuntu-24.04
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo add-apt-repository -y ppa:kisak/kisak-mesa
-          sudo apt-get update -y
-          sudo apt-get install -y build-essential mesa-vulkan-drivers \
-            libxcb-xinput0 libxcb-xinerama0 libxcb-cursor-dev libssl-dev
-
-      - name: Dawn Dependency
-        id: dawn-depends
-        run: |
-          sudo apt-get install -y libxrandr-dev libxinerama-dev libxcursor-dev mesa-common-dev libx11-xcb-dev libxi-dev
-          DAWN_VERSION="v20260317.182325"
-          DAWN_OWNER="google"
-          DAWN_REPO="dawn"
-          DAWN_ASSET_NAME="Dawn-18eb229ef5f707c1464cc581252e7603c73a3ef0-ubuntu-latest-Release"
-          echo "Fetching release asset from https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          curl -L -o artifact.tar.gz \
-            "https://github.com/google/dawn/releases/download/${DAWN_VERSION}/${DAWN_ASSET_NAME}.tar.gz"
-          mkdir dawn
-          tar -xvf artifact.tar.gz -C dawn --strip-components=1
-
-      - name: Build
-        id: cmake_build
-        run: |
-          export Dawn_DIR=dawn/lib64/cmake/Dawn
-          cmake -B build \
-            -DGGML_WEBGPU=ON
-          time cmake --build build --config Release -j $(nproc)
-
-      - name: Test
-        id: cmake_test
-        run: |
-          cd build
-          # This is using llvmpipe and runs slower than other backends
-          # test-backend-ops is too slow on llvmpipe, skip it
-          ctest -L main -E test-backend-ops --verbose --timeout 900
-
-  ubuntu-wasm:
-    strategy:
-      matrix:
-        include:
-          - build: 'x64'
-            os: ubuntu-24.04
-          - build: 'arm64'
-            os: ubuntu-24.04-arm
-
-    runs-on: ${{ matrix.os }}
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: webgpu-${{ matrix.os }}-wasm
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
-      - name: Install Emscripten
-        run: |
-          git clone https://github.com/emscripten-core/emsdk.git
-          cd emsdk
-          ./emsdk install latest
-          ./emsdk activate latest
-
-      - name: Fetch emdawnwebgpu
-        run: |
-          DAWN_TAG="v20260317.182325"
-          EMDAWN_PKG="emdawnwebgpu_pkg-${DAWN_TAG}.zip"
-          echo "Downloading ${EMDAWN_PKG}"
-          curl -L -o emdawn.zip \
-            "https://github.com/google/dawn/releases/download/${DAWN_TAG}/${EMDAWN_PKG}"
-          unzip emdawn.zip
-
-      - name: Build WASM WebGPU
-        run: |
-          source emsdk/emsdk_env.sh
-          emcmake cmake -B build-wasm \
-            -G "Ninja" \
-            -DCMAKE_BUILD_TYPE=Release \
-            -DGGML_WEBGPU=ON \
-            -DLLAMA_OPENSSL=OFF \
-            -DEMDAWNWEBGPU_DIR=emdawnwebgpu_pkg
-
-          time cmake --build build-wasm --config Release --target test-backend-ops -j $(nproc)

.github/workflows/check-vendor.yml

@@ -19,7 +19,7 @@ on:
 
 jobs:
   check-vendor:
-    runs-on: [self-hosted, fast]
+    runs-on: ubuntu-slim
 
     steps:
       - name: Checkout

.github/workflows/code-style.yml

@@ -1,51 +0,0 @@
-name: Code Style Checker
-
-on:
-  workflow_dispatch: # allows manual triggering
-  push:
-    branches:
-      - master
-  pull_request:
-    branches:
-      - master
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.head_ref && github.ref || github.run_id }}
-  cancel-in-progress: true
-
-jobs:
-  model-naming:
-    runs-on: [self-hosted, fast]
-    steps:
-      - uses: actions/checkout@v6
-      - name: Check model naming conventions
-        run: |
-          python3 - << 'EOF'
-          import re, os, sys
-
-          pairs = re.findall(
-              r'case\s+(LLM_ARCH_\w+)\s*:\s*\n\s+return new (llama_model_\w+)\s*\(',
-              open("src/llama-model.cpp").read())
-
-          errors = []
-          for arch, cls in pairs:
-              suffix  = arch[len("LLM_ARCH_"):]
-              csuffix = cls[len("llama_model_"):]
-              fname   = csuffix.replace("_", "-") + ".cpp"
-
-              if not re.fullmatch(r'[A-Z][A-Z0-9_]*',   suffix):
-                  errors.append(f"{arch}: suffix not upper snake case, example: LLM_ARCH_MY_MODEL")
-
-              if not re.fullmatch(r'[a-z][a-z0-9_]*', csuffix):
-                  errors.append(f"{arch}: class suffix not lower snake case, example: llama_model_my_model")
-
-              elif suffix.lower() != csuffix:
-                  errors.append(f"{arch}: arch/class name mismatch, expected class 'llama_model_{suffix.lower()}' but got '{cls}'")
-
-              elif not os.path.isfile(f"src/models/{fname}"):
-                  errors.append(f"{arch}: expects model file name to be src/models/{fname}, but not found")
-
-          if errors:
-              print('\n'.join(f"  - {e}" for e in errors)); sys.exit(1)
-          print(f"OK: {len(pairs)} mappings validated.")
-          EOF

.github/workflows/docker.yml

@@ -11,11 +11,6 @@ name: Publish Docker image
 
 on:
   workflow_dispatch: # allows manual triggering
-    inputs:
-      skip_s390x:
-        description: "Skip the s390x build target (useful for fast test runs that do not need the IBM Z runner)"
-        type: boolean
-        default: false
   schedule:
     # Rebuild daily rather than on every push because it is expensive
     - cron: '12 4 * * *'
@@ -69,8 +64,6 @@ jobs:
       - name: Generate build and merge matrices
         id: matrices
         shell: bash
-        env:
-          SKIP_S390X: ${{ inputs.skip_s390x || 'false' }}
         run: |
           set -euo pipefail
 
@@ -93,11 +86,6 @@ jobs:
           ]
           JSON
 
-          if [ "${SKIP_S390X}" = "true" ]; then
-            jq 'map(select(.platforms != "linux/s390x"))' build-matrix.json > build-matrix.json.tmp
-            mv build-matrix.json.tmp build-matrix.json
-          fi
-
           BUILD_MATRIX="$(jq -c . build-matrix.json)"
           MERGE_MATRIX="$(jq -c '
             reduce .[] as $entry ({}; .[$entry.tag] |= (
@@ -144,7 +132,6 @@ jobs:
         config: ${{ fromJSON(needs.prepare_matrices.outputs.build_matrix) }}
     steps:
       - name: Check out the repo
-        id: checkout
         uses: actions/checkout@v6
         with:
           fetch-depth: 0
@@ -200,10 +187,6 @@ jobs:
         env:
           GITHUB_REPOSITORY_OWNER: '${{ github.repository_owner }}'
 
-      - name: Get build date
-        id: build_date
-        run: echo "date=$(date -u +"%Y-%m-%dT%H:%M:%SZ")" >> $GITHUB_OUTPUT
-
       - name: Free Disk Space (Ubuntu)
         if: ${{ matrix.config.free_disk_space == true }}
         uses: ggml-org/free-disk-space@v1.3.1
@@ -228,26 +211,13 @@ jobs:
         with:
           context: .
           platforms: ${{ matrix.config.platforms }}
-          outputs: type=image,name=${{ steps.meta.outputs.image_repo }},push-by-digest=true,name-canonical=true,push=true,oci-mediatypes=true
+          outputs: type=image,name=${{ steps.meta.outputs.image_repo }},push-by-digest=true,name-canonical=true,push=true
           file: ${{ matrix.config.dockerfile }}
           target: full
           provenance: false
           build-args: |
-            BUILD_DATE=${{ steps.build_date.outputs.date }}
-            APP_VERSION=${{ needs.create_tag.outputs.source_tag }}
-            APP_REVISION=${{ steps.checkout.outputs.commit }}
-            IMAGE_URL=${{ github.server_url }}/${{ github.repository }}
-            IMAGE_SOURCE=${{ github.server_url }}/${{ github.repository }}
             ${{ matrix.config.ubuntu_version && format('UBUNTU_VERSION={0}', matrix.config.ubuntu_version) || '' }}
             ${{ matrix.config.cuda_version && format('CUDA_VERSION={0}', matrix.config.cuda_version) || '' }}
-          annotations: |
-            manifest:org.opencontainers.image.created=${{ steps.build_date.outputs.date }}
-            manifest:org.opencontainers.image.version=${{ needs.create_tag.outputs.source_tag }}
-            manifest:org.opencontainers.image.revision=${{ steps.checkout.outputs.commit }}
-            manifest:org.opencontainers.image.title=llama.cpp
-            manifest:org.opencontainers.image.description=LLM inference in C/C++
-            manifest:org.opencontainers.image.url=${{ github.server_url }}/${{ github.repository }}
-            manifest:org.opencontainers.image.source=${{ github.server_url }}/${{ github.repository }}
           # using github experimental cache
           #cache-from: type=gha
           #cache-to: type=gha,mode=max
@@ -265,26 +235,13 @@ jobs:
         with:
           context: .
           platforms: ${{ matrix.config.platforms }}
-          outputs: type=image,name=${{ steps.meta.outputs.image_repo }},push-by-digest=true,name-canonical=true,push=true,oci-mediatypes=true
+          outputs: type=image,name=${{ steps.meta.outputs.image_repo }},push-by-digest=true,name-canonical=true,push=true
           file: ${{ matrix.config.dockerfile }}
           target: light
           provenance: false
           build-args: |
-            BUILD_DATE=${{ steps.build_date.outputs.date }}
-            APP_VERSION=${{ needs.create_tag.outputs.source_tag }}
-            APP_REVISION=${{ steps.checkout.outputs.commit }}
-            IMAGE_URL=${{ github.server_url }}/${{ github.repository }}
-            IMAGE_SOURCE=${{ github.server_url }}/${{ github.repository }}
             ${{ matrix.config.ubuntu_version && format('UBUNTU_VERSION={0}', matrix.config.ubuntu_version) || '' }}
             ${{ matrix.config.cuda_version && format('CUDA_VERSION={0}', matrix.config.cuda_version) || '' }}
-          annotations: |
-            manifest:org.opencontainers.image.created=${{ steps.build_date.outputs.date }}
-            manifest:org.opencontainers.image.version=${{ needs.create_tag.outputs.source_tag }}
-            manifest:org.opencontainers.image.revision=${{ steps.checkout.outputs.commit }}
-            manifest:org.opencontainers.image.title=llama.cpp
-            manifest:org.opencontainers.image.description=LLM inference in C/C++
-            manifest:org.opencontainers.image.url=${{ github.server_url }}/${{ github.repository }}
-            manifest:org.opencontainers.image.source=${{ github.server_url }}/${{ github.repository }}
           # using github experimental cache
           #cache-from: type=gha
           #cache-to: type=gha,mode=max
@@ -302,26 +259,13 @@ jobs:
         with:
           context: .
           platforms: ${{ matrix.config.platforms }}
-          outputs: type=image,name=${{ steps.meta.outputs.image_repo }},push-by-digest=true,name-canonical=true,push=true,oci-mediatypes=true
+          outputs: type=image,name=${{ steps.meta.outputs.image_repo }},push-by-digest=true,name-canonical=true,push=true
           file: ${{ matrix.config.dockerfile }}
           target: server
           provenance: false
           build-args: |
-            BUILD_DATE=${{ steps.build_date.outputs.date }}
-            APP_VERSION=${{ needs.create_tag.outputs.source_tag }}
-            APP_REVISION=${{ steps.checkout.outputs.commit }}
-            IMAGE_URL=${{ github.server_url }}/${{ github.repository }}
-            IMAGE_SOURCE=${{ github.server_url }}/${{ github.repository }}
             ${{ matrix.config.ubuntu_version && format('UBUNTU_VERSION={0}', matrix.config.ubuntu_version) || '' }}
             ${{ matrix.config.cuda_version && format('CUDA_VERSION={0}', matrix.config.cuda_version) || '' }}
-          annotations: |
-            manifest:org.opencontainers.image.created=${{ steps.build_date.outputs.date }}
-            manifest:org.opencontainers.image.version=${{ needs.create_tag.outputs.source_tag }}
-            manifest:org.opencontainers.image.revision=${{ steps.checkout.outputs.commit }}
-            manifest:org.opencontainers.image.title=llama.cpp
-            manifest:org.opencontainers.image.description=LLM inference in C/C++
-            manifest:org.opencontainers.image.url=${{ github.server_url }}/${{ github.repository }}
-            manifest:org.opencontainers.image.source=${{ github.server_url }}/${{ github.repository }}
           # using github experimental cache
           #cache-from: type=gha
           #cache-to: type=gha,mode=max
@@ -386,15 +330,10 @@ jobs:
 
     steps:
       - name: Check out the repo
-        id: checkout
         uses: actions/checkout@v6
         with:
           fetch-depth: 0
 
-      - name: Get build date
-        id: build_date
-        run: echo "date=$(date -u +"%Y-%m-%dT%H:%M:%SZ")" >> $GITHUB_OUTPUT
-
       - name: Download digest metadata
         uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8
         with:
@@ -422,8 +361,6 @@ jobs:
           IMAGE_REPO="ghcr.io/${REPO_OWNER}/${REPO_NAME}"
           PREFIX="${IMAGE_REPO}:"
           SRC_TAG="${{ needs.create_tag.outputs.source_tag }}"
-          BUILD_DATE="${{ steps.build_date.outputs.date }}"
-          COMMIT_SHA="${{ steps.checkout.outputs.commit }}"
           TAGS="${{ matrix.config.tag }}"
           ARCHES="${{ matrix.config.arches }}"
           DIGEST_GLOB="/tmp/digests/*.tsv"
@@ -475,21 +412,11 @@ jobs:
                   refs+=("${IMAGE_REPO}@${digest}")
               done
 
-              local annotations=(
-                  --annotation "index:org.opencontainers.image.created=${BUILD_DATE}"
-                  --annotation "index:org.opencontainers.image.version=${SRC_TAG}"
-                  --annotation "index:org.opencontainers.image.revision=${COMMIT_SHA}"
-                  --annotation "index:org.opencontainers.image.title=llama.cpp"
-                  --annotation "index:org.opencontainers.image.description=LLM inference in C/C++"
-                  --annotation "index:org.opencontainers.image.url=${{ github.server_url }}/${{ github.repository }}"
-                  --annotation "index:org.opencontainers.image.source=${{ github.server_url }}/${{ github.repository }}"
-              )
-
               echo "Creating ${merged_tag} from ${refs[*]}"
-              docker buildx imagetools create "${annotations[@]}" --tag "${merged_tag}" "${refs[@]}"
+              docker buildx imagetools create --tag "${merged_tag}" "${refs[@]}"
 
               echo "Creating ${merged_versioned_tag} from ${refs[*]}"
-              docker buildx imagetools create "${annotations[@]}" --tag "${merged_versioned_tag}" "${refs[@]}"
+              docker buildx imagetools create --tag "${merged_versioned_tag}" "${refs[@]}"
           }
 
           for tag in $TAGS; do

.github/workflows/editorconfig.yml

@@ -2,6 +2,11 @@ name: EditorConfig Checker
 
 on:
   workflow_dispatch: # allows manual triggering
+    inputs:
+      create_release:
+        description: 'Create new release'
+        required: true
+        type: boolean
   push:
     branches:
       - master
@@ -15,7 +20,7 @@ concurrency:
 
 jobs:
   editorconfig:
-    runs-on: [self-hosted, fast]
+    runs-on: ubuntu-slim
     steps:
       - uses: actions/checkout@v6
       - uses: editorconfig-checker/action-editorconfig-checker@840e866d93b8e032123c23bac69dece044d4d84c # v2.2.0

.github/workflows/gguf-publish.yml

@@ -29,10 +29,10 @@ jobs:
       uses: actions/setup-python@v6
       with:
         python-version: '3.11'
-        pip-install: poetry==2.4.0
     - name: Install dependencies
       run: |
         cd gguf-py
+        python -m pip install poetry==2.3.2
         poetry install
 
     - name: Build package

.github/workflows/hip-quality-check.yml

@@ -28,9 +28,9 @@ concurrency:
 env:
   GGML_NLOOP: 3
   GGML_N_THREADS: 1
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
 
 jobs:
   ubuntu-22-hip-quality-check:
@@ -50,7 +50,7 @@ jobs:
       - name: ccache
         uses: ggml-org/ccache-action@v1.2.21
         with:
-          key: hip-quality-check-ubuntu-22.04
+          key: ubuntu-22-hip-quality-check
           evict-old-files: 1d
           save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
 

.github/workflows/pre-tokenizer-hashes.yml

@@ -3,16 +3,16 @@ name: Check Pre-Tokenizer Hashes
 on:
     push:
         paths:
-            - 'conversion/base.py'
+            - 'convert_hf_to_gguf.py'
             - 'convert_hf_to_gguf_update.py'
     pull_request:
         paths:
-            - 'conversion/base.py'
+            - 'convert_hf_to_gguf.py'
             - 'convert_hf_to_gguf_update.py'
 
 jobs:
     pre-tokenizer-hashes:
-        runs-on: [self-hosted, fast]
+        runs-on: ubuntu-slim
 
         steps:
         - name: Checkout repository
@@ -30,16 +30,16 @@ jobs:
 
         - name: Update pre-tokenizer hashes
           run: |
-              cp conversion/base.py /tmp
+              cp convert_hf_to_gguf.py /tmp
               .venv/bin/python convert_hf_to_gguf_update.py --check-missing
 
         - name: Check if committed pre-tokenizer hashes matches generated version
           run: |
-              if ! diff -q conversion/base.py /tmp/base.py; then
-                  echo "Model pre-tokenizer hashes (in conversion/base.py) do not match generated hashes (from convert_hf_to_gguf_update.py)."
-                  echo "To fix: run ./convert_hf_to_gguf_update.py and commit the updated conversion/base.py along with your changes"
+              if ! diff -q convert_hf_to_gguf.py /tmp/convert_hf_to_gguf.py; then
+                  echo "Model pre-tokenizer hashes (in convert_hf_to_gguf.py) do not match generated hashes (from convert_hf_to_gguf_update.py)."
+                  echo "To fix: run ./convert_hf_to_gguf_update.py and commit the updated convert_hf_to_gguf.py along with your changes"
                   echo "Differences found:"
-                  diff conversion/base.py /tmp/base.py || true
+                  diff convert_hf_to_gguf.py /tmp/convert_hf_to_gguf.py || true
                   exit 1
               fi
               echo "Model pre-tokenizer hashes are up to date."

.github/workflows/python-check-requirements.yml

@@ -20,7 +20,7 @@ concurrency:
 
 jobs:
   python-check-requirements:
-    runs-on: [self-hosted, CPU, fast]
+    runs-on: ubuntu-slim
     name: check-requirements
     steps:
       - name: Check out source repository

.github/workflows/python-lint.yml

@@ -21,7 +21,7 @@ concurrency:
 
 jobs:
   flake8-lint:
-    runs-on: [self-hosted, fast]
+    runs-on: ubuntu-slim
     name: Lint
     steps:
       - name: Check out source repository

.github/workflows/python-type-check.yml

@@ -22,7 +22,7 @@ concurrency:
 
 jobs:
   python-type-check:
-    runs-on: [self-hosted, fast]
+    runs-on: ubuntu-slim
     name: python type-check
     steps:
       - name: Check out source repository
@@ -31,7 +31,7 @@ jobs:
         uses: actions/setup-python@v6
         with:
           python-version: "3.11"
-          pip-install: -r requirements/requirements-all.txt ty==0.0.35
+          pip-install: -r requirements/requirements-all.txt ty==0.0.26
       # - name: Type-check with Pyright
       #   uses: jakebailey/pyright-action@v2
       #   with:

.github/workflows/server-sanitize.yml

@@ -26,10 +26,10 @@ on:
     ]
 
 env:
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-  LLAMA_ARG_LOG_VERBOSITY: 10
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_VERBOSITY: 10
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || github.run_id }}
@@ -37,7 +37,7 @@ concurrency:
 
 jobs:
   server:
-    runs-on: [self-hosted, CPU, Linux, llama-server]
+    runs-on: ubuntu-latest
 
     strategy:
       matrix:
@@ -46,19 +46,19 @@ jobs:
       fail-fast: false
 
     steps:
-      #- name: Dependencies
-      #  id: depends
-      #  run: |
-      #    sudo apt-get update
-      #    sudo apt-get -y install \
-      #      build-essential \
-      #      xxd \
-      #      git \
-      #      cmake \
-      #      curl \
-      #      wget \
-      #      language-pack-en \
-      #      libssl-dev
+      - name: Dependencies
+        id: depends
+        run: |
+          sudo apt-get update
+          sudo apt-get -y install \
+            build-essential \
+            xxd \
+            git \
+            cmake \
+            curl \
+            wget \
+            language-pack-en \
+            libssl-dev
 
       - name: Clone
         id: checkout
@@ -67,13 +67,6 @@ jobs:
           fetch-depth: 0
           ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
 
-      - name: Setup Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: "24"
-          cache: "npm"
-          cache-dependency-path: "tools/ui/package-lock.json"
-
       - name: Build
         id: cmake_build
         run: |

.github/workflows/server-self-hosted.yml

@@ -29,10 +29,10 @@ on:
     ]
 
 env:
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-  LLAMA_ARG_LOG_VERBOSITY: 10
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_VERBOSITY: 10
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || github.run_id }}
@@ -67,13 +67,6 @@ jobs:
           fetch-depth: 0
           ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
 
-      - name: Setup Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: "24"
-          cache: "npm"
-          cache-dependency-path: "tools/ui/package-lock.json"
-
       - name: Build
         id: cmake_build
         run: |
@@ -91,106 +84,42 @@ jobs:
           export ${{ matrix.extra_args }}
           pytest -v -x -m "not slow"
 
-  server-cuda:
-    runs-on: [self-hosted, llama-server, Linux, NVIDIA]
-
-    name: server-cuda (${{ matrix.wf_name }})
-    strategy:
-      matrix:
-        build_type: [Release]
-        wf_name: ["GPUx1"]
-        include:
-          - build_type: Release
-            extra_args: "LLAMA_ARG_BACKEND_SAMPLING=1"
-            wf_name:    "GPUx1, backend-sampling"
-      fail-fast: false
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-          ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build -DGGML_CUDA=ON -DGGML_SCHED_NO_REALLOC=ON
-          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
-
-      - name: Tests
-        id: server_integration_tests
-        if: ${{ (!matrix.disabled_on_pr || !github.event.pull_request) }}
-        run: |
-          cd tools/server/tests
-          python3 -m venv venv
-          source venv/bin/activate
-          pip install -r requirements.txt
-          export ${{ matrix.extra_args }}
-          pytest -v -x -m "not slow"
-
-  server-kleidiai:
-    runs-on: ah-ubuntu_22_04-c8g_8x
-
-    name: server-kleidiai (${{ matrix.wf_name }})
-    strategy:
-      matrix:
-        include:
-          - build_type: Release
-            extra_build_flags: "-DGGML_CPU_KLEIDIAI=ON"
-            extra_args: ""
-            wf_name:    "CPUx1, kleidiai"
-      fail-fast: false
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-          ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
-
-      - name: Dependencies
-        id: depends
-        run: |
-          set -euxo pipefail
-          sudo apt-get update
-          sudo DEBIAN_FRONTEND=noninteractive NEEDRESTART_MODE=a \
-          apt-get install -y \
-           build-essential \
-           libssl-dev \
-           python3-venv \
-           gpg \
-           wget \
-           time \
-           git-lfs
-
-          git lfs install
-
-          # install the latest cmake
-          sudo install -d /usr/share/keyrings
-          wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc \
-           | gpg --dearmor \
-           | sudo tee /usr/share/keyrings/kitware-archive-keyring.gpg >/dev/null
-          echo 'deb [signed-by=/usr/share/keyrings/kitware-archive-keyring.gpg] https://apt.kitware.com/ubuntu/ jammy main' \
-           | sudo tee /etc/apt/sources.list.d/kitware.list
-          sudo apt-get update
-          sudo apt-get install -y cmake
-
-      - name: Build
-        id: cmake_build
-        run: |
-          cmake -B build -DGGML_SCHED_NO_REALLOC=ON ${{ matrix.extra_build_flags }}
-          cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
-
-      - name: Tests
-        id: server_integration_tests
-        if: ${{ (!matrix.disabled_on_pr || !github.event.pull_request) }}
-        run: |
-          cd tools/server/tests
-          python3 -m venv venv
-          source venv/bin/activate
-          pip install -r requirements.txt
-          export ${{ matrix.extra_args }}
-          pytest -v -x -m "not slow"
+  # TODO: provision CUDA runner
+  #  server-cuda:
+  #    runs-on: [self-hosted, llama-server, Linux, NVIDIA]
+  #
+  #    name: server-cuda (${{ matrix.wf_name }})
+  #    strategy:
+  #      matrix:
+  #        build_type: [Release]
+  #        wf_name: ["GPUx1"]
+  #        include:
+  #          - build_type: Release
+  #            extra_args: "LLAMA_ARG_BACKEND_SAMPLING=1"
+  #            wf_name:    "GPUx1, backend-sampling"
+  #      fail-fast: false
+  #
+  #    steps:
+  #      - name: Clone
+  #        id: checkout
+  #        uses: actions/checkout@v6
+  #        with:
+  #          fetch-depth: 0
+  #          ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
+  #
+  #      - name: Build
+  #        id: cmake_build
+  #        run: |
+  #          cmake -B build -DGGML_SCHED_NO_REALLOC=ON
+  #          cmake --build build --config ${{ matrix.build_type }} -j $(sysctl -n hw.logicalcpu) --target llama-server
+  #
+  #      - name: Tests
+  #        id: server_integration_tests
+  #        if: ${{ (!matrix.disabled_on_pr || !github.event.pull_request) }}
+  #        run: |
+  #          cd tools/server/tests
+  #          python3 -m venv venv
+  #          source venv/bin/activate
+  #          pip install -r requirements.txt
+  #          export ${{ matrix.extra_args }}
+  #          pytest -v -x -m "not slow"

.github/workflows/server-webui.yml

@@ -1,7 +1,7 @@
-name: UI
+name: Server WebUI
 
 on:
-  workflow_dispatch:
+  workflow_dispatch: # allows manual triggering
     inputs:
       sha:
         description: 'Commit SHA1 to build'
@@ -11,39 +11,34 @@ on:
     branches:
       - master
     paths: [
-      '.github/workflows/ui.yml',
-      '.github/workflows/ui-build.yml',
-      'tools/ui/**.*',
-      'tools/server/tests/**.*'
+      '.github/workflows/server-webui.yml',
+      'tools/server/webui/**.*',
+      'tools/server/tests/**.*',
+      'tools/server/public/**'
     ]
   pull_request:
     types: [opened, synchronize, reopened]
     paths: [
-      '.github/workflows/ui.yml',
-      '.github/workflows/ui-build.yml',
-      'tools/ui/**.*',
-      'tools/server/tests/**.*'
+      '.github/workflows/server-webui.yml',
+      'tools/server/webui/**.*',
+      'tools/server/tests/**.*',
+      'tools/server/public/**'
     ]
 
 env:
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-  LLAMA_ARG_LOG_VERBOSITY: 10
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_VERBOSITY: 10
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || github.run_id }}
   cancel-in-progress: true
 
 jobs:
-  ui-build:
-    name: Build static output
-    uses: ./.github/workflows/ui-build.yml
-
-  ui-checks:
-    name: Checks
-    needs: ui-build
-    runs-on: ubuntu-latest
+  webui-check:
+    name: WebUI Checks
+    runs-on: ${{ 'ubuntu-24.04-arm' || 'ubuntu-24.04' }}
     continue-on-error: true
     steps:
       - name: Checkout code
@@ -56,89 +51,58 @@ jobs:
         id: node
         uses: actions/setup-node@v6
         with:
-          node-version: "24"
+          node-version: "22"
           cache: "npm"
-          cache-dependency-path: "tools/ui/package-lock.json"
+          cache-dependency-path: "tools/server/webui/package-lock.json"
 
       - name: Install dependencies
         id: setup
         if: ${{ steps.node.conclusion == 'success' }}
         run: npm ci
-        working-directory: tools/ui
+        working-directory: tools/server/webui
 
       - name: Run type checking
         if: ${{ always() && steps.setup.conclusion == 'success' }}
         run: npm run check
-        working-directory: tools/ui
+        working-directory: tools/server/webui
 
       - name: Run linting
         if: ${{ always() && steps.setup.conclusion == 'success' }}
         run: npm run lint
-        working-directory: tools/ui
-
-      - name: Install Playwright browsers
-        id: playwright
-        if: ${{ always() && steps.setup.conclusion == 'success' }}
-        run: npx playwright install --with-deps
-        working-directory: tools/ui
-
-      - name: Run Client tests
-        if: ${{ always() && steps.playwright.conclusion == 'success' }}
-        run: npm run test:client
-        working-directory: tools/ui
-
-      - name: Run Unit tests
-        if: ${{ always() && steps.playwright.conclusion == 'success' }}
-        run: npm run test:unit
-        working-directory: tools/ui
-
-  e2e-tests:
-    name: E2E Tests
-    needs: ui-build
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-          ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
-
-      - name: Setup Node.js
-        id: node
-        uses: actions/setup-node@v6
-        with:
-          node-version: "24"
-          cache: "npm"
-          cache-dependency-path: "tools/ui/package-lock.json"
-
-      - name: Install dependencies
-        id: setup
-        if: ${{ steps.node.conclusion == 'success' }}
-        run: npm ci
-        working-directory: tools/ui
+        working-directory: tools/server/webui
 
       - name: Build application
         if: ${{ always() && steps.setup.conclusion == 'success' }}
         run: npm run build
-        working-directory: tools/ui
+        working-directory: tools/server/webui
 
       - name: Install Playwright browsers
         id: playwright
         if: ${{ always() && steps.setup.conclusion == 'success' }}
         run: npx playwright install --with-deps
-        working-directory: tools/ui
+        working-directory: tools/server/webui
 
       - name: Build Storybook
         if: ${{ always() && steps.playwright.conclusion == 'success' }}
         run: npm run build-storybook
-        working-directory: tools/ui
+        working-directory: tools/server/webui
+
+      - name: Run Client tests
+        if: ${{ always() && steps.playwright.conclusion == 'success' }}
+        run: npm run test:client
+        working-directory: tools/server/webui
+
+      - name: Run Unit tests
+        if: ${{ always() && steps.playwright.conclusion == 'success' }}
+        run: npm run test:unit
+        working-directory: tools/server/webui
 
       - name: Run UI tests
         if: ${{ always() && steps.playwright.conclusion == 'success' }}
         run: npm run test:ui -- --testTimeout=60000
-        working-directory: tools/ui
+        working-directory: tools/server/webui
 
       - name: Run E2E tests
         if: ${{ always() && steps.playwright.conclusion == 'success' }}
         run: npm run test:e2e
-        working-directory: tools/ui
+        working-directory: tools/server/webui

.github/workflows/server.yml

@@ -44,20 +44,20 @@ on:
     ]
 
 env:
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-  LLAMA_ARG_LOG_VERBOSITY: 10
+  LLAMA_LOG_COLORS: 1
+  LLAMA_LOG_PREFIX: 1
+  LLAMA_LOG_TIMESTAMPS: 1
+  LLAMA_LOG_VERBOSITY: 10
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || github.run_id }}
   cancel-in-progress: true
 
 jobs:
-  ubuntu:
-    runs-on: ubuntu-24.04
+  server:
+    runs-on: ubuntu-latest
 
-    name: ubuntu (${{ matrix.wf_name }})
+    name: server (${{ matrix.wf_name }})
     strategy:
       matrix:
         build_type: [Release]
@@ -93,17 +93,11 @@ jobs:
           fetch-depth: 0
           ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
 
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: server-ubuntu-24.04-x64
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
       - name: Build
         id: cmake_build
         run: |
           cmake -B build \
+            -DLLAMA_BUILD_BORINGSSL=ON \
             -DGGML_SCHED_NO_REALLOC=ON
           cmake --build build --config ${{ matrix.build_type }} -j $(nproc) --target llama-server
 
@@ -130,8 +124,8 @@ jobs:
           export ${{ matrix.extra_args }}
           SLOW_TESTS=1 pytest -v -x
 
-  windows:
-    runs-on: windows-2025
+  server-windows:
+    runs-on: windows-2022
 
     steps:
       - name: Clone
@@ -141,24 +135,11 @@ jobs:
           fetch-depth: 0
           ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
 
-      - name: ccache
-        uses: ggml-org/ccache-action@v1.2.21
-        with:
-          key: server-windows-2025-x64
-          evict-old-files: 1d
-          save: ${{ github.event_name == 'push' && github.ref == 'refs/heads/master' }}
-
       - name: Build
         id: cmake_build
-        shell: cmd
         run: |
-          cmake -B build -G "Ninja Multi-Config" ^
-            -DCMAKE_TOOLCHAIN_FILE=cmake/x64-windows-llvm.cmake ^
-            -DCMAKE_BUILD_TYPE=Release ^
-            -DLLAMA_BUILD_BORINGSSL=ON ^
-            -DGGML_SCHED_NO_REALLOC=ON
-          set /A NINJA_JOBS=%NUMBER_OF_PROCESSORS%-1
-          cmake --build build --config Release -j %NINJA_JOBS% --target llama-server
+          cmake -B build -DLLAMA_BUILD_BORINGSSL=ON -DGGML_SCHED_NO_REALLOC=ON
+          cmake --build build --config Release -j ${env:NUMBER_OF_PROCESSORS} --target llama-server
 
       - name: Python setup
         id: setup_python

.github/workflows/ui-build-self-hosted.yml

@@ -1,43 +0,0 @@
-name: UI Build (self-hosted)
-
-on:
-  workflow_call:
-
-jobs:
-  build:
-    runs-on: [self-hosted, fast]
-    env:
-      BRANCH_NAME: ${{ github.head_ref || github.ref_name }}
-
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v6
-
-      - name: Setup Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: "24"
-          cache: "npm"
-          cache-dependency-path: "tools/ui/package-lock.json"
-
-      - name: Install dependencies
-        run: npm ci
-        working-directory: tools/ui
-
-      - name: Build application
-        run: npm run build
-        working-directory: tools/ui
-
-      - name: Generate checksums
-        run: |
-          cd tools/ui/dist
-          for f in *; do
-            sha256sum "$f" | awk '{print $1, $2}' >> checksums.txt
-          done
-
-      - name: Upload built UI
-        uses: actions/upload-artifact@v6
-        with:
-          name: ui-build
-          path: tools/ui/dist/
-          retention-days: 1

.github/workflows/ui-build.yml

@@ -1,43 +0,0 @@
-name: UI Build
-
-on:
-  workflow_call:
-
-jobs:
-  build:
-    runs-on: ubuntu-slim
-    env:
-      BRANCH_NAME: ${{ github.head_ref || github.ref_name }}
-
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v6
-
-      - name: Setup Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: "24"
-          cache: "npm"
-          cache-dependency-path: "tools/ui/package-lock.json"
-
-      - name: Install dependencies
-        run: npm ci
-        working-directory: tools/ui
-
-      - name: Build application
-        run: npm run build
-        working-directory: tools/ui
-
-      - name: Generate checksums
-        run: |
-          cd tools/ui/dist
-          for f in *; do
-            sha256sum "$f" | awk '{print $1, $2}' >> checksums.txt
-          done
-
-      - name: Upload built UI
-        uses: actions/upload-artifact@v6
-        with:
-          name: ui-build
-          path: tools/ui/dist/
-          retention-days: 1

.github/workflows/ui-publish.yml

@@ -1,70 +0,0 @@
-name: UI Publish
-
-on:
-  workflow_call:
-    inputs:
-      version_tag:
-        description: 'Version tag to publish under (e.g., b1234)'
-        required: true
-        type: string
-    secrets:
-      hf_token:
-        description: 'Hugging Face token with write access'
-        required: true
-
-jobs:
-  build:
-    name: Build static output
-    uses: ./.github/workflows/ui-build.yml
-
-  publish:
-    name: Publish UI Static Output
-    needs: build
-    runs-on: ubuntu-slim
-
-    permissions:
-      contents: read
-
-    env:
-      HF_BUCKET_NAME: ${{ vars.HF_BUCKET_UI_STATIC_OUTPUT }}
-
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 1
-
-      - name: Download UI build artifact
-        uses: actions/download-artifact@v7
-        with:
-          name: ui-build
-          path: tools/ui/dist/
-
-      - name: Install Hugging Face Hub CLI
-        run: pip install -U huggingface_hub
-
-      - name: Authenticate with Hugging Face
-        run: hf auth login --token ${{ secrets.hf_token }}
-
-      - name: Sync built files to Hugging Face bucket (version tag)
-        run: |
-          # Upload the built files to the Hugging Face bucket under the release version
-          hf buckets sync tools/ui/dist hf://buckets/ggml-org/${{ env.HF_BUCKET_NAME }}/${{ inputs.version_tag }} --delete --quiet
-
-      - name: Sync built files to Hugging Face bucket (latest)
-        run: |
-          # Also upload to the 'latest' directory for fallback downloads
-          hf buckets sync tools/ui/dist hf://buckets/ggml-org/${{ env.HF_BUCKET_NAME }}/latest --delete --quiet
-
-      - name: Verify upload
-        run: |
-          # List the files in the bucket to verify the upload
-          hf buckets list hf://buckets/ggml-org/${{ env.HF_BUCKET_NAME }}/${{ inputs.version_tag }} -R -h
-
-      - name: Clean up root-level files
-        run: |
-          # Clean up any old root-level files from previous non-versioned deployments
-          hf buckets rm ggml-org/${{ env.HF_BUCKET_NAME }}/index.html --yes 2>/dev/null || true
-          hf buckets rm ggml-org/${{ env.HF_BUCKET_NAME }}/bundle.js --yes 2>/dev/null || true
-          hf buckets rm ggml-org/${{ env.HF_BUCKET_NAME }}/bundle.css --yes 2>/dev/null || true
-          hf buckets rm ggml-org/${{ env.HF_BUCKET_NAME }}/loading.html --yes 2>/dev/null || true

.github/workflows/ui-self-hosted.yml

@@ -1,118 +0,0 @@
-name: UI (self-hosted)
-
-# these are the same as ui.yml, but with self-hosted runners
-# the runners come with pre-installed Playwright browsers version: 1.56.1
-# the jobs are much lighter because they don't need to install node and playwright browsers
-
-on:
-  workflow_dispatch:
-    inputs:
-      sha:
-        description: 'Commit SHA1 to build'
-        required: false
-        type: string
-  push:
-    branches:
-      - master
-    paths: [
-      '.github/workflows/ui-self-hosted.yml',
-      '.github/workflows/ui-build-self-hosted.yml',
-      'tools/ui/**.*',
-      'tools/server/tests/**.*'
-    ]
-  pull_request:
-    types: [opened, synchronize, reopened]
-    paths: [
-      '.github/workflows/ui-self-hosted.yml',
-      '.github/workflows/ui-build-self-hosted.yml',
-      'tools/ui/**.*',
-      'tools/server/tests/**.*'
-    ]
-
-env:
-  LLAMA_ARG_LOG_COLORS: 1
-  LLAMA_ARG_LOG_PREFIX: 1
-  LLAMA_ARG_LOG_TIMESTAMPS: 1
-  LLAMA_ARG_LOG_VERBOSITY: 10
-
-concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.head_ref || github.run_id }}
-  cancel-in-progress: true
-
-jobs:
-  ui-build:
-    name: Build static output
-    uses: ./.github/workflows/ui-build-self-hosted.yml
-
-  ui-checks:
-    name: Checks
-    needs: ui-build
-    runs-on: [self-hosted, PLAYWRIGHT]
-    continue-on-error: true
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-          ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
-
-      - name: Install dependencies
-        id: setup
-        run: npm ci
-        working-directory: tools/ui
-
-      - name: Run type checking
-        if: ${{ always() && steps.setup.conclusion == 'success' }}
-        run: npm run check
-        working-directory: tools/ui
-
-      - name: Run linting
-        if: ${{ always() && steps.setup.conclusion == 'success' }}
-        run: npm run lint
-        working-directory: tools/ui
-
-      - name: Run Client tests
-        if: ${{ always() }}
-        run: npm run test:client
-        working-directory: tools/ui
-
-      - name: Run Unit tests
-        if: ${{ always() }}
-        run: npm run test:unit
-        working-directory: tools/ui
-
-  e2e-tests:
-    name: E2E Tests
-    needs: ui-build
-    runs-on: [self-hosted, PLAYWRIGHT]
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-          ref: ${{ github.event.inputs.sha || github.event.pull_request.head.sha || github.sha || github.head_ref || github.ref_name }}
-
-      - name: Install dependencies
-        id: setup
-        run: npm ci
-        working-directory: tools/ui
-
-      - name: Build application
-        if: ${{ always() && steps.setup.conclusion == 'success' }}
-        run: npm run build
-        working-directory: tools/ui
-
-      - name: Build Storybook
-        if: ${{ always() }}
-        run: npm run build-storybook
-        working-directory: tools/ui
-
-      - name: Run UI tests
-        if: ${{ always() }}
-        run: npm run test:ui -- --testTimeout=60000
-        working-directory: tools/ui
-
-      - name: Run E2E tests
-        if: ${{ always() }}
-        run: npm run test:e2e
-        working-directory: tools/ui

.github/workflows/update-ops-docs.yml

@@ -3,20 +3,18 @@ name: Update Operations Documentation
 on:
     push:
         paths:
-            - '.github/workflows/update-ops-docs.yml'
             - 'docs/ops.md'
             - 'docs/ops/**'
             - 'scripts/create_ops_docs.py'
     pull_request:
         paths:
-            - '.github/workflows/update-ops-docs.yml'
             - 'docs/ops.md'
             - 'docs/ops/**'
             - 'scripts/create_ops_docs.py'
 
 jobs:
     update-ops-docs:
-        runs-on: [self-hosted, fast, ARM64]
+        runs-on: ubuntu-slim
 
         steps:
         - name: Checkout repository

.gitignore

@@ -34,6 +34,7 @@
 /.vscode/
 /nppBackup
 
+
 # Coverage
 
 /gcovr-report/
@@ -73,7 +74,6 @@
 !/models/templates
 
 # Zig
-
 /zig-out/
 /zig-cache/
 
@@ -92,12 +92,11 @@
 !/examples/sycl/*.bat
 !/examples/sycl/*.sh
 
-# Server Web UI temporary files (+ legacy directory)
-
+# Server Web UI temporary files
 /tools/server/webui/node_modules
 /tools/server/webui/dist
-/tools/ui/node_modules
-/tools/ui/dist
+# we no longer use gz for index.html
+/tools/server/public/index.html.gz
 
 # Python
 
@@ -105,16 +104,11 @@
 __pycache__/
 */poetry.lock
 poetry.toml
-poetry.lock
-uv.lock
 
 # Nix
-
-flake.lock
 /result
 
 # Test binaries
-
 /tests/test-backend-ops
 /tests/test-double-float
 /tests/test-grad0
@@ -130,7 +124,6 @@ flake.lock
 /tests/test-tokenizer-1-spm
 
 # Scripts
-
 !/scripts/install-oneapi.bat
 
 # Generated by scripts
@@ -139,24 +132,16 @@ flake.lock
 /wikitext-2-raw/
 
 # Test models for lora adapters
-
 /lora-tests
 
 # Local scripts
-
 /run-vim.sh
 /run-chat.sh
 /run-spec.sh
 /.ccache/
 
 # IDE
-
 /*.code-workspace
 /.windsurf/
 # emscripten
 a.out.*
-
-# AGENTS
-
-AGENTS.local.md
-.pi/SYSTEM.md

.pi/gg/SYSTEM.md

@@ -1,37 +0,0 @@
-You are a coding agent. Here are some very important rules that you must follow:
-
-General:
-- Be very precise and concise when writing code, comments, explanations, etc.
-- PR and commit titles format: `<module> : <title>`. Lookup recents for examples
-- Don't try to build or run the code unless you are explicitly asked to do so
-- Use the `gh` CLI tool when querying PRs, issues, or other GitHub resources
-
-Coding:
-- When in doubt, always refer to the CONTRIBUTING.md file of the project
-- When referencing issues or PRs in comments, use the format:
-  - C/C++ code: `// ref: <url>`
-  - Other (CMake, etc.): `# ref: <url>`
-
-Pull requests (PRs):
-- New branch names are prefixed with "gg/"
-- Before opening a pull request, ask the user to confirm the description
-- When creating a pull request, look for the repository's PR template and follow it
-- For the AI usage disclosure section, write "YES. llama.cpp + pi + [MODEL]"
-- Ask the user to tell you what model was used and write it in place of [MODEL]
-- Always create the pull requests in draft mode
-
-Commits:
-- On every commit that you make, include a "Assisted-by: llama.cpp:local pi" tag
-- Do not explicitly set the git author in commits - rely on the default git config
-- Always use `--no-gpg-sign` when committing
-- Never `git push` without explicit confirmation from the user
-
-Resources (read on demand):
-- [CONTRIBUTING.md](CONTRIBUTING.md)
-- [Build documentation](docs/build.md)
-- [Server usage documentation](tools/server/README.md)
-- [Server development documentation](tools/server/README-dev.md)
-- [PEG parser](docs/development/parsing.md)
-- [Auto parser](docs/autoparser.md)
-- [Jinja engine](common/jinja/README.md)
-- [PR template](.github/pull_request_template.md)

CMakeLists.txt

@@ -104,16 +104,13 @@ option(LLAMA_SANITIZE_UNDEFINED "llama: enable undefined sanitizer" OFF)
 option(LLAMA_BUILD_COMMON "llama: build common utils library" ${LLAMA_STANDALONE})
 
 # extra artifacts
-option(LLAMA_BUILD_TESTS     "llama: build tests"                                                                ${LLAMA_STANDALONE})
-option(LLAMA_BUILD_TOOLS     "llama: build tools"                                                                ${LLAMA_STANDALONE})
-option(LLAMA_BUILD_EXAMPLES  "llama: build examples"                                                             ${LLAMA_STANDALONE})
-option(LLAMA_BUILD_SERVER    "llama: build server example"                                                       ${LLAMA_STANDALONE})
-option(LLAMA_BUILD_APP       "llama: build the unified binary"                                                   ${LLAMA_STANDALONE})
-option(LLAMA_BUILD_UI        "llama: build the embedded Web UI for server"                                       ON)
-option(LLAMA_USE_PREBUILT_UI "llama: use prebuilt UI from HF Bucket when available (requires LLAMA_BUILD_UI=ON)" ON)
-
-option(LLAMA_TOOLS_INSTALL "llama: install tools" ${LLAMA_TOOLS_INSTALL_DEFAULT})
-option(LLAMA_TESTS_INSTALL "llama: install tests" ON)
+option(LLAMA_BUILD_TESTS    "llama: build tests"          ${LLAMA_STANDALONE})
+option(LLAMA_BUILD_TOOLS    "llama: build tools"          ${LLAMA_STANDALONE})
+option(LLAMA_BUILD_EXAMPLES "llama: build examples"       ${LLAMA_STANDALONE})
+option(LLAMA_BUILD_SERVER   "llama: build server example" ${LLAMA_STANDALONE})
+option(LLAMA_BUILD_WEBUI    "llama: build the embedded Web UI for server"  ON)
+option(LLAMA_TOOLS_INSTALL  "llama: install tools"        ${LLAMA_TOOLS_INSTALL_DEFAULT})
+option(LLAMA_TESTS_INSTALL  "llama: install tests"        ON)
 
 # 3rd party libs
 option(LLAMA_OPENSSL    "llama: use openssl to support HTTPS" ON)
@@ -218,8 +215,17 @@ if (LLAMA_BUILD_COMMON AND LLAMA_BUILD_TOOLS)
     add_subdirectory(tools)
 endif()
 
-if (LLAMA_BUILD_APP)
-    add_subdirectory(app)
+# Automatically add all files from the 'licenses' directory
+file(GLOB EXTRA_LICENSES "${CMAKE_SOURCE_DIR}/licenses/LICENSE-*")
+
+foreach(FILE_PATH ${EXTRA_LICENSES})
+    get_filename_component(FILE_NAME "${FILE_PATH}" NAME)
+    string(REGEX REPLACE "^LICENSE-" "" NAME "${FILE_NAME}")
+    license_add_file("${NAME}" "${FILE_PATH}")
+endforeach()
+
+if (LLAMA_BUILD_COMMON)
+    license_generate(llama-common)
 endif()
 
 #
@@ -264,6 +270,18 @@ install(FILES ${CMAKE_CURRENT_BINARY_DIR}/llama-config.cmake
               ${CMAKE_CURRENT_BINARY_DIR}/llama-version.cmake
         DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/llama)
 
+install(
+    FILES convert_hf_to_gguf.py
+    PERMISSIONS
+        OWNER_READ
+        OWNER_WRITE
+        OWNER_EXECUTE
+        GROUP_READ
+        GROUP_EXECUTE
+        WORLD_READ
+        WORLD_EXECUTE
+    DESTINATION ${CMAKE_INSTALL_BINDIR})
+
 configure_file(cmake/llama.pc.in
         "${CMAKE_CURRENT_BINARY_DIR}/llama.pc"
         @ONLY)

CODEOWNERS

@@ -15,7 +15,7 @@
 # ggml-org/llama-common     : ggerganov, aldehir, angt, danbev, ngxson, pwilkin
 # ggml-org/llama-mtmd       : ngxson
 # ggml-org/llama-server     : ggerganov, ngxson, allozaur, angt, ServeurpersoCom
-# ggml-org/llama-ui           : allozaur
+# ggml-org/llama-webui      : allozaur
 
 /.devops/*.Dockerfile                   @ngxson
 /.github/actions/                       @ggml-org/ci
@@ -23,10 +23,8 @@
 /ci/                                    @ggerganov
 /cmake/                                 @ggerganov
 /common/                                @ggml-org/llama-common
-/common/fit.*                           @JohannesGaessler
 /common/jinja/                          @CISC
 /common/ngram-map.*                     @srogmann
-/conversion/                            @CISC
 /convert_*.py                           @CISC
 /docs/backend/snapdragon/               @ggml-org/ggml-hexagon
 /examples/batched.swift/                @ggerganov
@@ -49,34 +47,33 @@
 /examples/parallel/                     @ggerganov
 /examples/passkey/                      @ggerganov
 /examples/retrieval/                    @ggerganov
+/examples/save-load-state/              @ggerganov
 /examples/speculative-simple/           @ggerganov
 /examples/speculative/                  @ggerganov
 /ggml/cmake/                            @ggerganov
 /ggml/include/                          @ggerganov
-/ggml/src/ggml-backend-meta.cpp         @JohannesGaessler
 /ggml/src/ggml-cann/                    @ggml-org/ggml-cann
 /ggml/src/ggml-common.h                 @ggerganov
 /ggml/src/ggml-cpu/                     @ggerganov
 /ggml/src/ggml-cpu/spacemit/            @alex-spacemit
 /ggml/src/ggml-cuda/                    @ggml-org/ggml-cuda
-/ggml/src/ggml-cuda/vendors/hip.h       @IMbackK
 /ggml/src/ggml-cuda/fattn-wmma*         @IMbackK
-/ggml/src/ggml-hexagon/                 @ggml-org/ggml-hexagon
 /ggml/src/ggml-hip/                     @IMbackK
+/ggml/src/ggml-cuda/vendors/hip.h       @IMbackK
 /ggml/src/ggml-impl.h                   @ggerganov
 /ggml/src/ggml-metal/                   @ggml-org/ggml-metal
 /ggml/src/ggml-opencl/                  @ggml-org/ggml-opencl
-/ggml/src/ggml-openvino/                @cavusmustafa @wine99
+/ggml/src/ggml-hexagon/                 @ggml-org/ggml-hexagon
 /ggml/src/ggml-opt.cpp                  @JohannesGaessler
 /ggml/src/ggml-quants.*                 @ggerganov
 /ggml/src/ggml-rpc/                     @ggml-org/ggml-rpc
 /ggml/src/ggml-sycl/                    @ggml-org/ggml-sycl
 /ggml/src/ggml-threading.*              @ggerganov
-/ggml/src/ggml-virtgpu/                 @kpouget
 /ggml/src/ggml-vulkan/                  @ggml-org/ggml-vulkan
+/ggml/src/ggml-virtgpu/                 @kpouget
 /ggml/src/ggml-webgpu/                  @ggml-org/ggml-webgpu
 /ggml/src/ggml-zdnn/                    @ggml-org/ggml-zdnn @Andreas-Krebbel @AlekseiNikiforovIBM
-/ggml/src/ggml-zendnn/                  @avinashcpandey @Jiten1parmar @z-vishal
+/ggml/src/ggml-openvino/                @cavusmustafa @wine99
 /ggml/src/ggml.c                        @ggerganov
 /ggml/src/ggml.cpp                      @ggerganov
 /ggml/src/gguf.cpp                      @JohannesGaessler @Green-Sky
@@ -107,7 +104,7 @@
 /tools/rpc/                             @ggml-org/ggml-rpc
 /tools/server/*                         @ggml-org/llama-server # no subdir
 /tools/server/tests/                    @ggml-org/llama-server
-/tools/ui/                              @ggml-org/llama-ui
+/tools/server/webui/                    @ggml-org/llama-webui
 /tools/tokenize/                        @ggerganov
 /tools/tts/                             @ggerganov
 /vendor/                                @ggerganov

CONTRIBUTING.md

@@ -46,9 +46,7 @@ Before submitting your PR:
     - provide KL divergence data calculated vs. the FP16/BF16 (whichever is the native precision) version for both the new type as well as types of similar size
     - provide [performance data](https://github.com/ggml-org/llama.cpp/tree/master/tools/llama-bench) for the new type in comparison to types of similar size on pure CPU
 - Consider allowing write access to your branch for faster reviews, as reviewers can push commits directly
-- If you are a new contributor
-    - Limit your open PRs to 1
-    - Do not submit trivial fixes (e.g. typos, formatting changes)
+- If you are a new contributor, limit your open PRs to 1.
 
 After submitting your PR:
 - Expect requests for modifications to ensure the code meets llama.cpp's standards for quality and long-term maintainability
@@ -63,7 +61,6 @@ After submitting your PR:
 - Optionally pick a `<module>` from here: https://github.com/ggml-org/llama.cpp/wiki/Modules
 - Let other maintainers merge their own PRs
 - When merging a PR, make sure you have a good understanding of the changes
-- If a PR does not warrant a new release, add `[no release]` in the squashed commit to spare CI resources
 - Be mindful of maintenance: most of the work going into a feature happens after the PR is merged. If the PR author is not committed to contribute long-term, someone else needs to take responsibility (you)
 
 Maintainers reserve the right to decline review or close pull requests for any reason, without any questions, particularly under any of the following conditions:

README.md

@@ -27,7 +27,6 @@ LLM inference in C/C++
 - Vim/Neovim plugin for FIM completions: https://github.com/ggml-org/llama.vim
 - Hugging Face Inference Endpoints now support GGUF out of the box! https://github.com/ggml-org/llama.cpp/discussions/9669
 - Hugging Face GGUF editor: [discussion](https://github.com/ggml-org/llama.cpp/discussions/9268) | [tool](https://huggingface.co/spaces/CISCai/gguf-editor)
-- WebGPU support is now available in the browser, see a blog/demo introducing it [here](https://reeselevine.github.io/llamas-on-the-web/).
 
 ----
 
@@ -173,7 +172,6 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - JavaScript/Wasm (works in browser): [tangledgroup/llama-cpp-wasm](https://github.com/tangledgroup/llama-cpp-wasm)
 - Typescript/Wasm (nicer API, available on npm): [ngxson/wllama](https://github.com/ngxson/wllama)
 - Ruby: [yoshoku/llama_cpp.rb](https://github.com/yoshoku/llama_cpp.rb)
-- Ruby: [docusealco/rllama](https://github.com/docusealco/rllama)
 - Rust (more features): [edgenai/llama_cpp-rs](https://github.com/edgenai/llama_cpp-rs)
 - Rust (nicer API): [mdrokz/rust-llama.cpp](https://github.com/mdrokz/rust-llama.cpp)
 - Rust (more direct bindings): [utilityai/llama-cpp-rs](https://github.com/utilityai/llama-cpp-rs)
@@ -281,7 +279,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 | [Metal](docs/build.md#metal-build) | Apple Silicon |
 | [BLAS](docs/build.md#blas-build) | All |
 | [BLIS](docs/backend/BLIS.md) | All |
-| [SYCL](docs/backend/SYCL.md) | Intel GPU |
+| [SYCL](docs/backend/SYCL.md) | Intel and Nvidia GPU |
 | [OpenVINO [In Progress]](docs/backend/OPENVINO.md) | Intel CPUs, GPUs, and NPUs |
 | [MUSA](docs/build.md#musa) | Moore Threads GPU |
 | [CUDA](docs/build.md#cuda) | Nvidia GPU |
@@ -291,7 +289,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 | [CANN](docs/build.md#cann) | Ascend NPU |
 | [OpenCL](docs/backend/OPENCL.md) | Adreno GPU |
 | [IBM zDNN](docs/backend/zDNN.md) | IBM Z & LinuxONE |
-| [WebGPU](docs/build.md#webgpu) | All |
+| [WebGPU [In Progress]](docs/build.md#webgpu) | All |
 | [RPC](https://github.com/ggml-org/llama.cpp/tree/master/tools/rpc) | All |
 | [Hexagon [In Progress]](docs/backend/snapdragon/README.md) | Snapdragon |
 | [VirtGPU](docs/backend/VirtGPU.md) | VirtGPU APIR |
@@ -531,7 +529,6 @@ To learn more about model quantization, [read this documentation](tools/quantize
 - [How to build](docs/build.md)
 - [Running on Docker](docs/docker.md)
 - [Build on Android](docs/android.md)
-- [Multi-GPU usage](docs/multi-gpu.md)
 - [Performance troubleshooting](docs/development/token_generation_performance_tips.md)
 - [GGML tips & tricks](https://github.com/ggml-org/llama.cpp/wiki/GGML-Tips-&-Tricks)
 

app/CMakeLists.txt

@@ -1,31 +0,0 @@
-set(TARGET llama-app)
-
-add_executable(${TARGET} llama.cpp)
-set_target_properties(${TARGET} PROPERTIES OUTPUT_NAME llama)
-
-target_link_libraries(${TARGET} PRIVATE
-    llama-server-impl
-    llama-cli-impl
-    llama-completion-impl
-    llama-bench-impl
-    llama-batched-bench-impl
-    llama-fit-params-impl
-    llama-quantize-impl
-    llama-perplexity-impl
-)
-target_compile_features(${TARGET} PRIVATE cxx_std_17)
-
-# Automatically add all files from the 'licenses' directory
-file(GLOB EXTRA_LICENSES "${CMAKE_SOURCE_DIR}/licenses/LICENSE-*")
-
-foreach(FILE_PATH ${EXTRA_LICENSES})
-    get_filename_component(FILE_NAME "${FILE_PATH}" NAME)
-    string(REGEX REPLACE "^LICENSE-" "" NAME "${FILE_NAME}")
-    license_add_file("${NAME}" "${FILE_PATH}")
-endforeach()
-
-license_generate(${TARGET})
-
-if(LLAMA_TOOLS_INSTALL)
-    install(TARGETS ${TARGET} RUNTIME)
-endif()

app/llama.cpp

@@ -1,127 +0,0 @@
-#include "build-info.h"
-
-#include <cstdio>
-#include <cstdlib>
-#include <string>
-#include <vector>
-
-// embedded data generated by cmake
-extern const char * LICENSES[];
-
-// visible
-int llama_server(int argc, char ** argv);
-int llama_cli(int argc, char ** argv);
-
-// hidden
-int llama_completion(int argc, char ** argv);
-int llama_bench(int argc, char ** argv);
-int llama_batched_bench(int argc, char ** argv);
-int llama_fit_params(int argc, char ** argv);
-int llama_quantize(int argc, char ** argv);
-int llama_perplexity(int argc, char ** argv);
-
-// hands the update over to the install script, which downloads and swaps the binary
-static int llama_update(int argc, char ** argv) {
-    (void) argc;
-    (void) argv;
-
-#if defined(_WIN32)
-    return system("powershell -NoProfile -ExecutionPolicy Bypass -Command \"irm https://llama.app/install.ps1 | iex\"");
-#else
-    return system("curl -fsSL https://llama.app/install.sh | sh");
-#endif
-}
-
-static const char * progname;
-
-static int help(int argc, char ** argv);
-static int version(int argc, char ** argv);
-static int licenses(int argc, char ** argv);
-
-struct command {
-    const char * name;
-    const char * desc;
-    std::vector<std::string> aliases;
-    bool hidden;
-    int (*func)(int, char **);
-};
-
-static const command cmds[] = {
-    {"serve",         "HTTP API server",                                    {"server"},   false, llama_server       },
-    {"cli",           "Command-line interactive interface",                 {"client"},   false, llama_cli          },
-    {"update",        "Update llama to the latest release",                 {},           false, llama_update       },
-    {"completion",    "Text completion",                                    {"complete"}, true,  llama_completion   },
-    {"bench",         "Benchmark prompt processing and text generation",    {},           true,  llama_bench        },
-    {"batched-bench", "Benchmark batched decoding performance",             {},           true,  llama_batched_bench},
-    {"fit-params",    "Compute parameters to fit a model in device memory", {},           true,  llama_fit_params   },
-    {"quantize",      "Quantize a model",                                   {},           true,  llama_quantize     },
-    {"perplexity",    "Compute model perplexity and KL divergence",         {},           true,  llama_perplexity   },
-    {"version",       "Show version",                                       {},           false, version            },
-    {"licenses",      "Show third-party licenses",                          {"credits"},  false, licenses           },
-    {"help",          "Show available commands",                            {},           false, help               },
-};
-
-static int version(int argc, char ** argv) {
-    printf("%s\n", llama_build_info());
-    return 0;
-}
-
-static int licenses(int argc, char ** argv) {
-    for (int i = 0; LICENSES[i]; ++i) {
-        printf("%s\n", LICENSES[i]);
-    }
-    return 0;
-}
-
-static int help(int argc, char ** argv) {
-    const bool show_all = argc >= 2 && std::string(argv[1]) == "all";
-
-    printf("Usage: %s <command> [options]\n\nAvailable commands:\n", progname);
-
-    for (const auto & cmd : cmds) {
-        if (show_all || !cmd.hidden) {
-            printf("  %-15s %s\n", cmd.name, cmd.desc);
-        }
-    }
-    printf("\n");
-
-    if (!show_all) {
-        printf("Run '%s help all' to show additional commands.\n", progname);
-    }
-    printf("Run '%s <command> --help' for command-specific usage.\n", progname);
-
-    return 0;
-}
-
-static bool matches(const std::string & arg, const command & cmd) {
-    if (arg == cmd.name) {
-        return true;
-    }
-    for (const auto & alias : cmd.aliases) {
-        if (arg == alias) {
-            return true;
-        }
-    }
-    return false;
-}
-
-int main(int argc, char ** argv) {
-    progname = argv[0];
-
-    const std::string arg = argc >= 2 ? argv[1] : "help";
-
-    for (const auto & cmd : cmds) {
-        if (matches(arg, cmd)) {
-            // keep cmd.name so the router's child processes re-invoke correctly
-#ifdef _WIN32
-            _putenv_s("LLAMA_APP_CMD", cmd.name);
-#else
-            setenv("LLAMA_APP_CMD", cmd.name, 1);
-#endif
-            return cmd.func(argc - 1, argv + 1);
-        }
-    }
-
-    fprintf(stderr, "error: unknown command '%s'\n", arg.c_str());
-    return 1;
-}

build-xcframework.sh

@@ -7,7 +7,6 @@ VISIONOS_MIN_OS_VERSION=1.0
 TVOS_MIN_OS_VERSION=16.4
 
 BUILD_SHARED_LIBS=OFF
-LLAMA_BUILD_APP=OFF
 LLAMA_BUILD_EXAMPLES=OFF
 LLAMA_BUILD_TOOLS=OFF
 LLAMA_BUILD_TESTS=OFF
@@ -32,7 +31,6 @@ COMMON_CMAKE_ARGS=(
     -DCMAKE_XCODE_ATTRIBUTE_STRIP_INSTALLED_PRODUCT=NO
     -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
     -DBUILD_SHARED_LIBS=${BUILD_SHARED_LIBS}
-    -DLLAMA_BUILD_APP=${LLAMA_BUILD_APP}
     -DLLAMA_BUILD_EXAMPLES=${LLAMA_BUILD_EXAMPLES}
     -DLLAMA_BUILD_TOOLS=${LLAMA_BUILD_TOOLS}
     -DLLAMA_BUILD_TESTS=${LLAMA_BUILD_TESTS}
@@ -416,7 +414,7 @@ cmake -B build-ios-sim -G Xcode \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
     -S .
-cmake --build build-ios-sim --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
+cmake --build build-ios-sim --config Release -- -quiet
 
 echo "Building for iOS devices..."
 cmake -B build-ios-device -G Xcode \
@@ -430,7 +428,7 @@ cmake -B build-ios-device -G Xcode \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
     -S .
-cmake --build build-ios-device --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
+cmake --build build-ios-device --config Release -- -quiet
 
 echo "Building for macOS..."
 cmake -B build-macos -G Xcode \
@@ -441,7 +439,7 @@ cmake -B build-macos -G Xcode \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
     -S .
-cmake --build build-macos --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
+cmake --build build-macos --config Release -- -quiet
 
 echo "Building for visionOS..."
 cmake -B build-visionos -G Xcode \
@@ -456,7 +454,7 @@ cmake -B build-visionos -G Xcode \
     -DLLAMA_OPENSSL=OFF \
     -DLLAMA_BUILD_SERVER=OFF \
     -S .
-cmake --build build-visionos --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
+cmake --build build-visionos --config Release -- -quiet
 
 echo "Building for visionOS simulator..."
 cmake -B build-visionos-sim -G Xcode \
@@ -471,7 +469,7 @@ cmake -B build-visionos-sim -G Xcode \
     -DLLAMA_OPENSSL=OFF \
     -DLLAMA_BUILD_SERVER=OFF \
     -S .
-cmake --build build-visionos-sim --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
+cmake --build build-visionos-sim --config Release -- -quiet
 
 # Add tvOS builds (might need the same u_int definitions as watchOS and visionOS)
 echo "Building for tvOS simulator..."
@@ -487,7 +485,7 @@ cmake -B build-tvos-sim -G Xcode \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
     -S .
-cmake --build build-tvos-sim --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
+cmake --build build-tvos-sim --config Release -- -quiet
 
 echo "Building for tvOS devices..."
 cmake -B build-tvos-device -G Xcode \
@@ -502,7 +500,7 @@ cmake -B build-tvos-device -G Xcode \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
     -DLLAMA_OPENSSL=OFF \
     -S .
-cmake --build build-tvos-device --config Release -j $(sysctl -n hw.logicalcpu) -- -quiet
+cmake --build build-tvos-device --config Release -- -quiet
 
 # Setup frameworks and copy binaries and headers
 echo "Setting up framework structures..."

ci/run.sh

@@ -66,8 +66,6 @@ fi
 
 if [ ! -z ${GG_BUILD_METAL} ]; then
     CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_METAL=ON"
-else
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_METAL=OFF"
 fi
 
 if [ ! -z ${GG_BUILD_CUDA} ]; then
@@ -116,12 +114,9 @@ fi
 if [ ! -z ${GG_BUILD_VULKAN} ]; then
     CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_VULKAN=1"
 
+    # if on Mac, disable METAL
     if [[ "$OSTYPE" == "darwin"* ]]; then
-        MACOS_RUNNER_CUSTOM_VULKAN_CMAKE_LOCATION="/usr/local/lib/cmake/vulkan"
-        MACOS_RUNNER_CUSTOM_SPIRV_HEADERS_LOCATION="${MACOS_RUNNER_CUSTOM_VULKAN_CMAKE_LOCATION}/SPIRV-Headers/SPIRV-HeadersConfig.cmake"
-        if [[ -f "${MACOS_RUNNER_CUSTOM_SPIRV_HEADERS_LOCATION}" || -h "${MACOS_RUNNER_CUSTOM_SPIRV_HEADERS_LOCATION}" ]]; then
-            CMAKE_EXTRA="${CMAKE_EXTRA} -DSPIRV-Headers_DIR=${MACOS_RUNNER_CUSTOM_VULKAN_CMAKE_LOCATION}/SPIRV-Headers"
-        fi
+        CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_METAL=OFF -DGGML_BLAS=OFF"
     fi
 
     # Build shared libs on Windows
@@ -132,7 +127,7 @@ if [ ! -z ${GG_BUILD_VULKAN} ]; then
 fi
 
 if [ ! -z ${GG_BUILD_WEBGPU} ]; then
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_WEBGPU=1"
+    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_WEBGPU=1 -DGGML_METAL=OFF -DGGML_BLAS=OFF"
 
     if [ ! -z "${GG_BUILD_WEBGPU_DAWN_PREFIX}" ]; then
         if [ -z "${CMAKE_PREFIX_PATH}" ]; then
@@ -166,8 +161,6 @@ fi
 
 if [ ! -z ${GG_BUILD_BLAS} ]; then
     CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_BLAS=ON -DGGML_BLAS_VENDOR=${GG_BUILD_BLAS_VENDOR:-OpenBLAS}"
-else
-    CMAKE_EXTRA="${CMAKE_EXTRA} -DGGML_BLAS=OFF"
 fi
 
 if [ ! -z ${GG_BUILD_OPENVINO} ]; then
@@ -239,7 +232,7 @@ function gg_run_ctest_debug {
     (cmake -G "${CMAKE_GENERATOR}" -DCMAKE_BUILD_TYPE=Debug ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time cmake --build . --config Debug -j$(nproc)) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    (time ctest -C Debug --output-on-failure -L main -E "test-opt|test-backend-ops|test-llama-archs" ${CTEST_EXTRA}) 2>&1 | tee -a $OUT/${ci}-ctest.log
+    (time ctest -C Debug --output-on-failure -L main -E "test-opt|test-backend-ops" ${CTEST_EXTRA}) 2>&1 | tee -a $OUT/${ci}-ctest.log
 
     set +e
 }
@@ -462,10 +455,10 @@ function gg_run_qwen3_0_6b {
 
     (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -ngl 99 -c 1024 -b 512 --chunks 2 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
 
-    (time ./bin/test-save-load-state --model ${model_q4_0} -ngl 10 -c 1024 -fa off --no-op-offload) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
-    (time ./bin/test-save-load-state --model ${model_q4_0} -ngl 10 -c 1024 -fa on  --no-op-offload) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
-    (time ./bin/test-save-load-state --model ${model_q4_0} -ngl 99 -c 1024 -fa off                ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
-    (time ./bin/test-save-load-state --model ${model_q4_0} -ngl 99 -c 1024 -fa on                 ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 1024 -fa off --no-op-offload) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 1024 -fa on  --no-op-offload) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 1024 -fa off                ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 1024 -fa on                 ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
 
     function check_ppl {
         qnt="$1"
@@ -701,8 +694,8 @@ function gg_sum_test_backend_ops_cpu {
 
 ## main
 
-export LLAMA_ARG_LOG_PREFIX=1
-export LLAMA_ARG_LOG_TIMESTAMPS=1
+export LLAMA_LOG_PREFIX=1
+export LLAMA_LOG_TIMESTAMPS=1
 
 if [ -z ${GG_BUILD_LOW_PERF} ]; then
     # Create symlink: ./llama.cpp/models-mnt -> $MNT/models

cmake/llama-config.cmake.in

@@ -7,7 +7,7 @@ set(LLAMA_SHARED_LIB   @BUILD_SHARED_LIBS@)
 
 set_and_check(LLAMA_INCLUDE_DIR "@PACKAGE_LLAMA_INCLUDE_INSTALL_DIR@")
 set_and_check(LLAMA_LIB_DIR     "@PACKAGE_LLAMA_LIB_INSTALL_DIR@")
-set(LLAMA_BIN_DIR "@PACKAGE_LLAMA_BIN_INSTALL_DIR@")
+set_and_check(LLAMA_BIN_DIR     "@PACKAGE_LLAMA_BIN_INSTALL_DIR@")
 
 find_package(ggml REQUIRED HINTS ${LLAMA_LIB_DIR}/cmake)
 

cmake/riscv64-spacemit-linux-gnu-gcc.cmake

@@ -24,6 +24,6 @@ set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
 set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
 set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
 set(CMAKE_FIND_ROOT_PATH_MODE_PACKAGE ONLY)
-set(CMAKE_C_FLAGS "-march=rv64gcv_zfh_zvfh_zba_zicbop -mabi=lp64d -fno-tree-vectorize -fno-tree-loop-vectorize ${CMAKE_C_FLAGS}")
-set(CMAKE_CXX_FLAGS "-march=rv64gcv_zfh_zvfh_zba_zicbop -mabi=lp64d -fno-tree-vectorize -fno-tree-loop-vectorize ${CMAKE_CXX_FLAGS}")
+set(CMAKE_C_FLAGS "-march=rv64gcv_zfh_zba_zicbop -mabi=lp64d ${CMAKE_C_FLAGS}")
+set(CMAKE_CXX_FLAGS "-march=rv64gcv_zfh_zba_zicbop -mabi=lp64d ${CXX_FLAGS}")
 set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -latomic")

common/CMakeLists.txt

@@ -73,8 +73,6 @@ add_library(${TARGET}
     debug.h
     download.cpp
     download.h
-    fit.cpp
-    fit.h
     hf-cache.cpp
     hf-cache.h
     http.h

common/arg.h

@@ -25,8 +25,7 @@ struct common_arg {
     const char * value_hint_2 = nullptr; // for second arg value
     const char * env          = nullptr;
     std::string help;
-    bool is_sampling = false; // is current arg a sampling param?
-    bool is_spec = false; // is current arg a speculative decoding param?
+    bool is_sparam = false; // is current arg a sampling param?
     bool is_preset_only = false; // is current arg preset-only (not treated as CLI arg)
     void (*handler_void)   (common_params & params) = nullptr;
     void (*handler_string) (common_params & params, const std::string &) = nullptr;
@@ -75,8 +74,7 @@ struct common_arg {
     common_arg & set_examples(std::initializer_list<enum llama_example> examples);
     common_arg & set_excludes(std::initializer_list<enum llama_example> excludes);
     common_arg & set_env(const char * env);
-    common_arg & set_sampling();
-    common_arg & set_spec();
+    common_arg & set_sparam();
     common_arg & set_preset_only();
     bool in_example(enum llama_example ex);
     bool is_exclude(enum llama_example ex);
@@ -129,11 +127,5 @@ bool common_params_to_map(int argc, char ** argv, llama_example ex, std::map<com
 // see: https://github.com/ggml-org/llama.cpp/issues/18163
 void common_params_add_preset_options(std::vector<common_arg> & args);
 
-// populate model paths (main model, mmproj, etc) from -hf if necessary
-// return true if the model is ready to use
-// throw an exception if there is an error that prevents the model from being used (e.g. network error, model not found, etc)
-// if params.skip_download is true, no downloads will be attempted. return false if the model is invalid or missing (e.g. ETag check failed)
-bool common_params_handle_models(common_params & params, llama_example curr_ex);
-
 // initialize argument parser context - used by test-arg-parser and preset
 common_params_context common_params_parser_init(common_params & params, llama_example ex, void(*print_usage)(int, char **) = nullptr);

common/chat-auto-parser-generator.cpp

@@ -43,33 +43,11 @@ common_chat_params peg_generator::generate_parser(const common_chat_template &
                                                   const autoparser &              autoparser) {
     // Create the result structure
     common_chat_params data;
-    data.prompt            = common_chat_template_direct_apply(tmpl, inputs);
-    data.generation_prompt = common_chat_template_generation_prompt(tmpl, inputs);
-    data.format            = COMMON_CHAT_FORMAT_PEG_NATIVE;
-    data.preserved_tokens  = autoparser.preserved_tokens;
+    data.prompt           = common_chat_template_direct_apply(tmpl, inputs);
+    data.format           = COMMON_CHAT_FORMAT_PEG_NATIVE;
+    data.preserved_tokens = autoparser.preserved_tokens;
 
-    std::string parser_generation_prompt = data.generation_prompt;
-
-    if (inputs.continue_final_message != COMMON_CHAT_CONTINUATION_NONE && !inputs.continue_msg.empty()) {
-        // Build up generation prompt manually
-        const auto & msg = inputs.continue_msg;
-
-        if (!autoparser.reasoning.start.empty()) {
-            data.generation_prompt = data.generation_prompt.substr(0, data.generation_prompt.find(autoparser.reasoning.start));
-            data.generation_prompt += autoparser.reasoning.start + msg.reasoning_content;
-            if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
-                data.generation_prompt += autoparser.reasoning.end;
-            }
-        }
-
-        if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
-            data.generation_prompt += msg.render_content();
-        }
-
-        data.prompt += data.generation_prompt;
-    }
-
-    auto parser = autoparser.build_parser(inputs, parser_generation_prompt);
+    auto parser = autoparser.build_parser(inputs);
     data.parser = parser.save();
 
     // Build grammar if tools are present
@@ -109,7 +87,7 @@ common_chat_params peg_generator::generate_parser(const common_chat_template &
     return data;
 }
 
-common_peg_arena autoparser::build_parser(const generation_params & inputs, const std::string & generation_prompt) const {
+common_peg_arena autoparser::build_parser(const generation_params & inputs) const {
     if (!analysis_complete) {
         throw std::invalid_argument("Cannot call build_parser on autoparser without performing analysis first, call analyze_template(...)");
     }
@@ -143,7 +121,7 @@ common_peg_arena autoparser::build_parser(const generation_params & inputs, cons
         } else {
             parser = content.build_parser(ctx);
         }
-        return pure_content ? p.prefix(generation_prompt, reasoning.start) + parser : p.prefix(generation_prompt, reasoning.start) << parser;
+        return pure_content ? p.prefix(inputs.generation_prompt, reasoning.start) + parser : p.prefix(inputs.generation_prompt, reasoning.start) << parser;
     });
 }
 
@@ -158,10 +136,10 @@ common_peg_parser analyze_reasoning::build_parser(parser_build_context & ctx) co
         if (!end.empty()) {
             if (!start.empty()) {
                 // Standard tag-based: optional(<think>reasoning</think>)
-                return p.optional(p.optspace(start) + p.reasoning(p.until(trim_whitespace(end))) + p.optspace(end));
+                return p.optional(start + p.reasoning(p.until(end)) + end + p.space());
             }
             // Delimiter-style (empty start)
-            return p.optional(p.reasoning(p.until(trim_whitespace(end))) + p.optspace(end));
+            return p.optional(p.reasoning(p.until(end)) + end + p.space());
         }
     }
 
@@ -208,6 +186,7 @@ common_peg_parser analyze_tools::build_parser(parser_build_context & ctx) const
 common_peg_parser analyze_tools::build_tool_parser_json_native(parser_build_context & ctx) const {
     auto &       p           = ctx.p;
     const auto & inputs      = ctx.inputs;
+    bool         force_tools = inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED;
 
     // Build effective field names with dot notation if function_field is set
     std::string name_field = format.name_field;
@@ -246,7 +225,8 @@ common_peg_parser analyze_tools::build_tool_parser_json_native(parser_build_cont
         tool_start = format.per_call_start;
     }
 
-    return ctx.reasoning_parser + p.optional(p.content(p.until(tool_start))) + tools_parser + p.end();
+    return ctx.reasoning_parser + (force_tools ? p.eps() : p.optional(p.content(p.until(tool_start)))) + tools_parser +
+           p.end();
 }
 
 common_peg_parser analyze_tools::build_func_parser(common_chat_peg_builder & p, const std::string & name,
@@ -290,6 +270,7 @@ common_peg_parser analyze_tools::build_func_parser(common_chat_peg_builder & p,
 common_peg_parser analyze_tools::build_tool_parser_tag_json(parser_build_context & ctx) const {
     auto &       p           = ctx.p;
     const auto & inputs      = ctx.inputs;
+    bool         force_tools = inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED;
 
     common_peg_parser tool_choice = p.choice();
 
@@ -355,12 +336,14 @@ common_peg_parser analyze_tools::build_tool_parser_tag_json(parser_build_context
 
     std::string trigger_marker       = !format.section_start.empty() ? format.section_start : format.per_call_start;
     auto        content_before_tools = trigger_marker.empty() ? p.eps() : p.until(trigger_marker);
-    return ctx.reasoning_parser + p.optional(p.content(content_before_tools)) + tool_calls + p.end();
+    return ctx.reasoning_parser + (force_tools ? p.eps() : p.optional(p.content(content_before_tools))) + tool_calls +
+           p.end();
 }
 
 common_peg_parser analyze_tools::build_tool_parser_tag_tagged(parser_build_context & ctx) const {
     auto &       p           = ctx.p;
     const auto & inputs      = ctx.inputs;
+    bool         force_tools = inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_REQUIRED;
 
     auto until_suffix = p.rule("until-suffix", p.until(arguments.value_suffix));
 
@@ -391,7 +374,9 @@ common_peg_parser analyze_tools::build_tool_parser_tag_tagged(parser_build_conte
                                            arguments.name_suffix) +
                            arguments.value_prefix +
                            (schema_info.resolves_to_string(param_schema) ?
-                                p.tool_arg_string_value(until_suffix) :
+                                p.tool_arg_string_value(p.schema(until_suffix,
+                                                                 "tool-" + name + "-arg-" + param_name + "-schema",
+                                                                 param_schema, true)) :
                                 p.tool_arg_json_value(p.schema(
                                     p.json(), "tool-" + name + "-arg-" + param_name + "-schema", param_schema, false)) +
                                     p.space()) +
@@ -486,7 +471,8 @@ common_peg_parser analyze_tools::build_tool_parser_tag_tagged(parser_build_conte
 
     std::string trigger_marker       = !format.section_start.empty() ? format.section_start : format.per_call_start;
     auto        content_before_tools = trigger_marker.empty() ? p.eps() : p.until(trigger_marker);
-    return ctx.reasoning_parser + p.optional(p.content(content_before_tools)) + tool_calls + p.end();
+    return ctx.reasoning_parser + (force_tools ? p.eps() : p.optional(p.content(content_before_tools))) + tool_calls +
+           p.end();
 }
 
 }  // namespace autoparser

common/chat-auto-parser-helpers.cpp

@@ -310,8 +310,6 @@ std::vector<segment> prune_whitespace_segments(const std::vector<segment> & segm
 
 namespace autoparser {
 
-static const std::string ERR_TMPL = "#**ERROR**#";
-
 std::string apply_template(const common_chat_template & tmpl, const template_params & params) {
     generation_params tmpl_params;
     tmpl_params.messages              = params.messages;
@@ -328,7 +326,7 @@ std::string apply_template(const common_chat_template & tmpl, const template_par
         return common_chat_template_direct_apply(tmpl, tmpl_params);
     } catch (const std::exception & e) {
         LOG_DBG("Template application failed: %s\n", e.what());
-        return ERR_TMPL;
+        return "";
     }
 }
 
@@ -349,7 +347,7 @@ std::optional<compare_variants_result> compare_variants(
     std::string output_B = apply_template(tmpl, params_B);
 
     // Check for template application failures
-    if (output_A == ERR_TMPL || output_B == ERR_TMPL) {
+    if (output_A.empty() || output_B.empty()) {
         return std::nullopt;
     }
 

common/chat-auto-parser.h

@@ -60,21 +60,16 @@ struct generation_params {
     common_reasoning_format               reasoning_format    = COMMON_REASONING_FORMAT_AUTO;
     bool                                  stream              = true;
     std::string                           grammar;
-    bool                                  add_generation_prompt  = false;
-    common_chat_continuation              continue_final_message = COMMON_CHAT_CONTINUATION_NONE;
-    common_chat_msg                       continue_msg;
-    bool                                  enable_thinking        = true;
-    std::chrono::system_clock::time_point now                    = std::chrono::system_clock::now();
+    bool                                  add_generation_prompt = false;
+    bool                                  enable_thinking       = true;
+    std::chrono::system_clock::time_point now                   = std::chrono::system_clock::now();
+    std::string                           generation_prompt;
     json                                  extra_context;
     bool                                  add_bos       = false;
     bool                                  add_eos       = false;
     bool                                  is_inference  = true;
     bool                                  add_inference = false;
     bool                                  mark_input    = true;  // whether to mark input strings in the jinja context
-
-    bool has_continuation() const {
-        return continue_final_message != COMMON_CHAT_CONTINUATION_NONE && !continue_msg.empty();
-    }
 };
 
 // ============================================================================
@@ -377,8 +372,6 @@ struct analyze_tools : analyze_base {
 
 struct autoparser {
     jinja::caps          jinja_caps;
-    std::string          user_start;
-    std::string          assistant_start;
     analyze_reasoning    reasoning;
     analyze_content      content;
     analyze_tools        tools;
@@ -389,15 +382,11 @@ struct autoparser {
 
     autoparser() = default;
 
-    // Find the starting marker for the user message and assistant message
-    std::string detect_user_start_marker(const common_chat_template & tmpl);
-    std::string detect_assistant_start_marker(const common_chat_template & tmpl);
-
     // Run full differential analysis on a template
     void analyze_template(const common_chat_template & tmpl);
 
     // Build the PEG parser for this template
-    common_peg_arena build_parser(const generation_params & inputs, const std::string & generation_prompt) const;
+    common_peg_arena build_parser(const generation_params & inputs) const;
 
   private:
     // Collect tokens from entire analysis to preserve

common/chat-diff-analyzer.cpp

@@ -8,9 +8,6 @@
 #include "peg-parser.h"
 
 #include <algorithm>
-#include <cctype>
-#include <ostream>
-#include <sstream>
 
 #define ANSI_RESET  "\033[0m"
 #define ANSI_PURPLE "\033[1m\x1b[38;5;126m"
@@ -26,7 +23,6 @@ static const std::string FUN_SECOND = "SSS_SECOND_FUN_S";
 static const std::string ARG_FIRST = "AA_ARG_FST_AA";
 static const std::string ARG_SECOND = "BB_ARG_SND_BB";
 static const std::string USER_MSG = "U_USER_MSG Hello END_U";
-static const std::string USER_MSG_TWO = "V_USER_MSG Hello END_V";
 static const std::string ASSISTANT_MSG = "A_ASST_MSG I can help END_A";
 static const std::string THINKING_CONTENT = "REASON_PART I am thinking END_R";
 static const std::string CALL_ID_001 = "call00001";
@@ -75,7 +71,6 @@ static std::vector<std::function<void(const common_chat_template & tmpl, autopar
               analysis.content.end   = "<|END_OF_TURN_TOKEN|>";
               analysis.preserved_tokens.push_back("<|CHATBOT_TOKEN|>");
               analysis.preserved_tokens.push_back("<|END_OF_TURN_TOKEN|>");
-              analysis.user_start = "<|START_OF_TURN_TOKEN|><|USER_TOKEN|>";
               LOG_DBG(ANSI_ORANGE "[Patch: Cohere Command R+]\n" ANSI_RESET);
           }
       },
@@ -113,59 +108,7 @@ static std::vector<std::function<void(const common_chat_template & tmpl, autopar
               analysis.tools.function.close        = "```";
               LOG_DBG(ANSI_ORANGE "[Patch: DeepSeek-R1-Distill-Qwen]\n" ANSI_RESET);
           }
-      },
-      // Nemotron Nano v2
-      [](const common_chat_template & tmpl, autoparser & analysis) -> void {
-          if (tmpl.src.find("<SPECIAL_10>") != std::string::npos && tmpl.src.find("<SPECIAL_11>") != std::string::npos &&
-              tmpl.src.find("<SPECIAL_12>") != std::string::npos && tmpl.src.find("<TOOL_RESPONSE>") != std::string::npos) {
-
-              analysis.tools.format.mode           = tool_format::JSON_NATIVE;
-              analysis.tools.format.section_start  = "";
-              analysis.tools.format.section_end    = "";
-              analysis.tools.format.per_call_start = "<TOOLCALL>";
-              analysis.tools.format.per_call_end   = "</TOOLCALL>";
-              analysis.content.mode                = content_mode::PLAIN;
-              analysis.content.start               = "";
-              analysis.content.end                 = "";
-              analysis.reasoning.mode              = reasoning_mode::TAG_BASED;
-              analysis.reasoning.start             = "<think>\n\n";
-              analysis.reasoning.end               = "</think>";
-              analysis.assistant_start             = "<SPECIAL_11>Assistant";
-              analysis.user_start                  = "<SPECIAL_11>User";
-              analysis.preserved_tokens.clear();
-              analysis.preserved_tokens.push_back("<SPECIAL_12>");
-              analysis.preserved_tokens.push_back("<SPECIAL_11>");
-              analysis.preserved_tokens.push_back("</think>");
-              analysis.preserved_tokens.push_back("<TOOLCALL>");
-              analysis.preserved_tokens.push_back("</TOOLCALL>");
-              LOG_DBG(ANSI_ORANGE "[Patch: Nemotron Nano v2]\n" ANSI_RESET);
-          }
-      },
-      // Fireworks
-      [](const common_chat_template & tmpl, autoparser & analysis) -> void {
-          if (tmpl.src.find("{%- set system_prompt = '<|start_header_id|>' + 'system' + '<|end_header_id|>\\n\\n'"
-            " + message['content'] | trim + '\\n' + system_prompt_suffix + '<|eot_id|>' -%}") != std::string::npos) {
-              analysis.assistant_start             = "<|start_header_id|>assistant<|end_header_id|>";
-              analysis.user_start                  = "<|start_header_id|>user<|end_header_id|>";
-              LOG_DBG(ANSI_ORANGE "[Patch: Fireworks v2]\n" ANSI_RESET);
-          }
-      },
-      // Solar Open
-      [](const common_chat_template & tmpl, autoparser & analysis) -> void {
-          if (tmpl.src.find("<|begin|>assistant<|think|><|end|>") != std::string::npos) {
-              analysis.assistant_start             = "<|begin|>assistant";
-              LOG_DBG(ANSI_ORANGE "[Patch: Solar Open]\n" ANSI_RESET);
-          }
-      },
-      // Apriel 1.6
-      [](const common_chat_template & tmpl, autoparser & analysis) -> void {
-          if (tmpl.src.find("if not loop.last and '[BEGIN FINAL RESPONSE]' in asst_text") != std::string::npos) {
-              analysis.user_start                  = "<|begin_user|>";
-              analysis.assistant_start             = "<|begin_assistant|>";
-              LOG_DBG(ANSI_ORANGE "[Patch: Apriel 1.6]\n" ANSI_RESET);
-          }
-      },
-
+      }
     });
 
 // Common JSON structures
@@ -223,8 +166,6 @@ void autoparser::analyze_template(const common_chat_template & tmpl) {
     reasoning = analyze_reasoning(tmpl, jinja_caps.supports_tool_calls);
     content = analyze_content(tmpl, reasoning);
     tools = analyze_tools(jinja_caps.supports_tool_calls ? analyze_tools(tmpl, jinja_caps, reasoning) : analyze_tools());
-    assistant_start = detect_assistant_start_marker(tmpl);
-    user_start = detect_user_start_marker(tmpl);
     collect_preserved_tokens();
 
     for (auto & workaround : workarounds) {
@@ -232,8 +173,6 @@ void autoparser::analyze_template(const common_chat_template & tmpl) {
     }
 
     LOG_DBG("\n--- Reasoning & Content Structure ---\n");
-    LOG_DBG("user_msg_start: %s\n", user_start.c_str());
-    LOG_DBG("assistant_msg_start: %s\n", assistant_start.c_str());
     LOG_DBG("reasoning_mode: %s\n", mode_to_str(reasoning.mode).c_str());
     LOG_DBG("reasoning_start: '%s'\n", reasoning.start.c_str());
     LOG_DBG("reasoning_end: '%s'\n", reasoning.end.c_str());
@@ -306,120 +245,6 @@ void autoparser::collect_preserved_tokens() {
     add_token(tools.call_id.suffix);
 }
 
-std::string autoparser::detect_assistant_start_marker(const common_chat_template & tmpl) {
-    json user_msg = json{
-        { "role",    "user"   },
-        { "content", USER_MSG }
-    };
-
-    json assistant_no_reasoning = json{
-        { "role",    "assistant"   },
-        { "content", ASSISTANT_MSG }
-    };
-
-    template_params params;
-    params.messages              = json::array({ user_msg });
-    params.add_generation_prompt = false;
-    params.enable_thinking       = true;
-
-    auto comparison = compare_variants(
-        tmpl, params, [&](template_params & p) {
-            p.messages = json::array({ user_msg, assistant_no_reasoning });
-        }
-    );
-
-    if (!comparison) {
-        LOG_DBG(ANSI_ORANGE "%s: Template application failed, skipping assistant start detection\n" ANSI_RESET, __func__);
-        return "";
-    }
-
-    auto usermsg = comparison->diff.right;
-    if (usermsg.find(ASSISTANT_MSG) == std::string::npos) {
-        LOG_DBG(ANSI_ORANGE "%s: Did not find assistant message in assistant message block, skipping detection\n" ANSI_RESET, __func__);
-    }
-
-    auto ast_prefix = usermsg.substr(0, usermsg.find(ASSISTANT_MSG));
-    if (!reasoning.start.empty() && ast_prefix.find(trim_whitespace(reasoning.start)) != std::string::npos) {
-        ast_prefix = ast_prefix.substr(0, ast_prefix.find(trim_whitespace(reasoning.start)));
-    }
-    if (!reasoning.end.empty() && ast_prefix.find(trim_whitespace(reasoning.end)) != std::string::npos) {
-        ast_prefix = ast_prefix.substr(0, ast_prefix.find(trim_whitespace(reasoning.end)));
-    }
-    return trim_whitespace(ast_prefix);
-}
-
-std::string autoparser::detect_user_start_marker(const common_chat_template & tmpl) {
-    json user_msg = json{
-        { "role",    "user"   },
-        { "content", USER_MSG }
-    };
-
-    json assistant = json{
-        { "role",    "assistant"   },
-        { "content", ASSISTANT_MSG }
-    };
-
-    json user_msg_two = json{
-        { "role",    "user"       },
-        { "content", USER_MSG_TWO }
-    };
-
-    template_params params;
-    params.messages              = json::array({});
-    params.add_generation_prompt = false;
-    params.enable_thinking       = true;
-
-    auto comparison = compare_variants(
-        tmpl, params, [&](template_params & p) {
-            p.messages = json::array({ user_msg });
-        }
-    );
-
-    if (!comparison) {
-        LOG_DBG(ANSI_ORANGE "%s: Template application failed, unsupported empty messages? trying complex variant\n" ANSI_RESET, __func__);
-        params.messages = json::array({ user_msg_two, assistant });
-        comparison = compare_variants(
-            tmpl, params, [&](template_params & p) {
-                p.messages = json::array({ user_msg_two, assistant, user_msg });
-            }
-        );
-        if (!comparison) {
-            LOG_DBG(ANSI_ORANGE "%s: Template application failed for reserve variant, aborting\n" ANSI_RESET, __func__);
-            return "";
-        }
-    }
-
-    auto usermsg = comparison->diff.right;
-    if (usermsg.find(USER_MSG) == std::string::npos) {
-        LOG_DBG(ANSI_ORANGE "%s: Did not find user message in user message block, aborting detection\n" ANSI_RESET, __func__);
-    }
-
-    if (usermsg.find(ASSISTANT_MSG) != std::string::npos) {
-        usermsg = usermsg.substr(usermsg.find(ASSISTANT_MSG) + ASSISTANT_MSG.size());
-    }
-
-    auto candidate = usermsg.substr(0, usermsg.find(USER_MSG));
-    auto candidate_split = segmentize_markers(candidate);
-    std::stringstream result;
-    bool encountered_marker = false;
-    for (const auto & mrk : candidate_split) {
-        std::string lower_mrk = std::string(mrk.value);
-        std::transform(lower_mrk.begin(), lower_mrk.end(), lower_mrk.begin(),
-            [](unsigned char c) { return std::tolower(c); });
-        // heuristic to weed out potential end markers, but only at the start
-        if (mrk.type == segment_type::MARKER && !encountered_marker &&
-            (lower_mrk.find("end") != std::string::npos || lower_mrk.find("close") != std::string::npos)) {
-            continue;
-        }
-        if (mrk.type == segment_type::TEXT && !encountered_marker && trim_whitespace(mrk.value).empty()) {
-            continue;
-        }
-        encountered_marker |= mrk.type == segment_type::MARKER;
-        result << mrk.value;
-    }
-    return trim_whitespace(result.str());
-}
-
 analyze_reasoning::analyze_reasoning(const common_chat_template & tmpl, bool supports_tools)
     : analyze_base(tmpl) {
     LOG_DBG(ANSI_PURPLE "=== Starting differential analysis ===\n" ANSI_RESET);
@@ -471,7 +296,7 @@ void analyze_reasoning::compare_reasoning_presence() {
             return p.literal(reasoning_content) + p.space() + p.optional(p.tag("post", (p.marker() + p.space())) + p.rest());
         });
         auto parser_wrapped = build_tagged_peg_parser([&](common_peg_parser_builder &p) {
-            return p.tag("pre", p.marker() + p.space()) + p.literal(reasoning_content) + p.tag("post", (p.space() + p.marker() + p.space())) + p.rest();
+            return p.tag("pre", p.marker() + p.space()) + p.literal(reasoning_content) + p.space() + p.tag("post", (p.marker() + p.space())) + p.rest();
         });
         // try the more aggressive parse first, if it fails, fall back to the delimiter one
         auto result = parser_wrapped.parse_anywhere_and_extract(comparison->output_B);
@@ -481,11 +306,11 @@ void analyze_reasoning::compare_reasoning_presence() {
         if (result.result.success()) {
             if (!result.tags["pre"].empty() && !result.tags["post"].empty()) {
                 mode = reasoning_mode::TAG_BASED;
-                start = result.tags["pre"];
-                end   = result.tags["post"];
+                start = trim_leading_whitespace(result.tags["pre"]);
+                end   = trim_trailing_whitespace(result.tags["post"]);
             } else if (!result.tags["post"].empty()) {
                 mode = reasoning_mode::TAG_BASED;
-                end = result.tags["post"];
+                end = trim_trailing_whitespace(result.tags["post"]);
             }
         }
     }
@@ -517,7 +342,7 @@ void analyze_reasoning::compare_thinking_enabled() {
     if (left_trimmed.empty() && !diff.right.empty()) {
         if (!right_trimmed.empty() && string_ends_with(comparison->output_B, right_trimmed)) {
             if (start.empty()) {
-                start = diff.right;
+                start = trim_leading_whitespace(diff.right);
                 mode  = reasoning_mode::TAG_BASED;
             }
         }
@@ -528,7 +353,7 @@ void analyze_reasoning::compare_thinking_enabled() {
                 if (seg.size() >= 2 && seg[seg.size() - 1].value == left_trimmed && seg[seg.size() - 2].type == segment_type::MARKER) {
                     start = seg[seg.size() - 2].value;
                 }
-                end = diff.left;
+                end = trim_trailing_whitespace(diff.left);
                 mode = reasoning_mode::TAG_BASED;
             }
         }
@@ -620,14 +445,14 @@ void analyze_reasoning::compare_reasoning_scope() {
         auto result = parser_wrapped.parse_anywhere_and_extract(comparison->output_B);
         if (result.result.success()) {
             start = result.tags["pre"];
-            end = result.tags["post"];
+            end = trim_trailing_whitespace(result.tags["post"]);
         } else {
             auto parser_delimiter = build_tagged_peg_parser([&](common_peg_parser_builder &p) {
                 return p.literal(reasoning_content) + p.space() + p.optional(p.tag("post", (p.marker() + p.space())));
             });
             result = parser_delimiter.parse_anywhere_and_extract(comparison->output_B);
             if (result.result.success()) {
-                end = result.tags["post"];
+                end = trim_trailing_whitespace(result.tags["post"]);
             } else {
                 LOG_DBG(ANSI_ORANGE "%s: Unable to extract reasoning markers, falling back to reasoning = NONE\n" ANSI_RESET, __func__);
                 mode = reasoning_mode::NONE;

common/chat-peg-parser.cpp

@@ -358,7 +358,35 @@ void common_chat_peg_mapper::map(const common_peg_ast_node & node) {
             if (is_potential_container) {
                 value_content = normalize_container_value(value_content);
             }
-            value_to_add += value_content;
+
+            // Try to parse as JSON value (number, bool, null, object, array)
+            try {
+                ordered_json parsed = ordered_json::parse(value_content);
+                if (parsed.is_string()) {
+                    // Don't add closing quote yet (added by arg_close) for monotonic streaming
+                    std::string escaped = parsed.dump();
+                    if (!escaped.empty() && escaped.back() == '"') {
+                        escaped.pop_back();
+                    }
+                    value_to_add          = escaped;
+                    closing_quote_pending = true;
+                } else {
+                    // Non-string values: use raw content to preserve whitespace for monotonicity
+                    value_to_add = value_content;
+                }
+            } catch (...) {
+                if (node.is_partial && is_potential_container) {
+                    // Partial container: pass through the already-normalized content
+                    value_to_add = value_content;
+                } else {
+                    // Not valid JSON - treat as string value
+                    if (!closing_quote_pending) {
+                        value_to_add          = "\"";
+                        closing_quote_pending = true;
+                    }
+                    value_to_add += escape_json_string_inner(value_content);
+                }
+            }
         }
 
         args_target() += value_to_add;
@@ -785,33 +813,7 @@ common_peg_parser common_chat_peg_builder::prefix(const std::string & s, const s
     if (delimiter.empty()) {
         return literal(s);
     }
-    return literal(s.substr(0, s.find(delimiter)));
-}
-
-common_peg_parser common_chat_peg_builder::optspace(const std::string & tag) {
-    auto parser = eps();
-    size_t end_of_prefix_space = tag.size();
-    size_t start_of_suffix_space = tag.size();
-    for (size_t i = 0; i < tag.size(); i++) {
-        if (!std::isspace(tag[i])) {
-            end_of_prefix_space = i;
-            break;
-        }
-    }
-    for (size_t i = tag.size(); i > 0; i--) {
-        if (!std::isspace(tag[i - 1])) {
-            start_of_suffix_space = i;
-            break;
-        }
-    }
-    for (size_t i = 0; i < end_of_prefix_space; i++) {
-        parser += optional(literal(std::string(1, tag[i])));
-    }
-    parser += literal(tag.substr(end_of_prefix_space, start_of_suffix_space - end_of_prefix_space));
-    for (size_t i = start_of_suffix_space; i < tag.size(); i++) {
-        parser += optional(literal(std::string(1, tag[i])));
-    }
-    return parser;
+    return literal(s.substr(0, s.rfind(delimiter)));
 }
 
 common_peg_parser common_chat_peg_builder::standard_json_tools(

common/chat-peg-parser.h

@@ -90,15 +90,12 @@ class common_chat_peg_builder : public common_peg_parser_builder {
 
     // Use for schema-declared string types - won't be treated as potential JSON container
     common_peg_parser tool_arg_string_value(const common_peg_parser & p) { return tag(TOOL_ARG_STRING_VALUE, p); }
-    common_peg_parser tool_arg_json_value(const common_peg_parser & p) { return tag(TOOL_ARG_VALUE, p); }
+    common_peg_parser tool_arg_json_value(const common_peg_parser & p) { return atomic(tag(TOOL_ARG_VALUE, p)); }
 
 
     // Return a parser that parses the prefix of a string, up to a given delimiter.
     common_peg_parser prefix(const std::string & s, const std::string & delimiter = {});
 
-    // Return a parser that parses all elements of tag, but leading and trailing spaces are optional
-    common_peg_parser optspace(const std::string & tag);
-
     // Legacy-compatible helper for building standard JSON tool calls
     // Used by tests and manual parsers
     // name_key/args_key: JSON key names for function name and arguments

common/chat.cpp

@@ -70,65 +70,6 @@ static bool has_content_or_tool_calls(const common_chat_msg & msg) {
     return !msg.content.empty() || !msg.tool_calls.empty();
 }
 
-std::string common_chat_msg::render_content(const std::string & delimiter) const {
-    if (!content.empty() && !content_parts.empty()) {
-        throw std::runtime_error("Cannot specify both content and content_parts");
-    }
-    if (!content.empty()) {
-        return content;
-    }
-
-    std::string text;
-    for (const auto & part : content_parts) {
-        if (part.type == "text") {
-            if (!text.empty()) {
-                text += delimiter;
-            }
-            text += part.text;
-        }
-    }
-    return text;
-}
-
-std::vector<common_chat_msg_span> common_chat_split_by_role(const std::string & prompt, const std::vector<common_chat_msg_delimiter> & delims) {
-    if (delims.empty() || prompt.empty()) {
-        return {};
-    }
-
-    auto parser = build_peg_parser([&](common_peg_parser_builder & p) {
-        std::vector<std::string>       all_delims;
-        std::vector<common_peg_parser> tagged_messages;
-
-        all_delims.reserve(delims.size());
-        tagged_messages.reserve(delims.size());
-        for (const auto & d : delims) {
-            all_delims.push_back(d.delimiter);
-        }
-
-        auto any_delim = p.until_one_of(all_delims);
-        for (const auto & d : delims) {
-            tagged_messages.push_back(p.tag(d.role, p.literal(d.delimiter) + any_delim));
-        }
-
-        return any_delim + p.zero_or_more(p.choice(tagged_messages)) + p.end();
-    });
-
-    common_peg_parse_context ctx(prompt);
-    const auto result = parser.parse(ctx);
-    if (!result.success()) {
-        return {};
-    }
-
-    std::vector<common_chat_msg_span> spans;
-    ctx.ast.visit(result, [&](const common_peg_ast_node & node) {
-        if (!node.tag.empty()) {
-            spans.push_back({ node.tag, node.start, node.end - node.start });
-        }
-    });
-
-    return spans;
-}
-
 json common_chat_msg::to_json_oaicompat(bool concat_typed_text) const {
     if (!content.empty() && !content_parts.empty()) {
         throw std::runtime_error("Cannot specify both content and content_parts");
@@ -139,7 +80,7 @@ json common_chat_msg::to_json_oaicompat(bool concat_typed_text) const {
     if (!content.empty()) {
         jmsg["content"] = content;
     } else if (!content_parts.empty()) {
-        if (concat_typed_text || contains_media()) {
+        if (concat_typed_text) {
             std::string text;
             bool last_was_media_marker = false;
             // join parts with newline, do not add newline before or after media markers
@@ -456,25 +397,6 @@ json common_chat_msgs_to_json_oaicompat(const std::vector<common_chat_msg> & msg
     return render_message_to_json(msgs, c);
 }
 
-json common_chat_tools_to_json_oaicompat(const std::vector<common_chat_tool> & tools) {
-    if (tools.empty()) {
-        return json();
-    }
-
-    auto result = json::array();
-    for (const auto & tool : tools) {
-        result.push_back({
-            { "type",     "function" },
-            { "function", {
-                { "name", tool.name },
-                { "description", tool.description },
-                { "parameters", json::parse(tool.parameters) },
-            }},
-        });
-    }
-    return result;
-}
-
 std::vector<common_chat_tool> common_chat_tools_parse_oaicompat(const json & tools) {
     std::vector<common_chat_tool> result;
 
@@ -510,20 +432,54 @@ std::vector<common_chat_tool> common_chat_tools_parse_oaicompat(const json & too
     return result;
 }
 
-common_chat_continuation common_chat_continuation_parse(const nlohmann::ordered_json & value) {
-    if (value.is_boolean() && value.get<bool>()) {
-        return COMMON_CHAT_CONTINUATION_AUTO;
+json common_chat_tools_to_json_oaicompat(const std::vector<common_chat_tool> & tools) {
+    if (tools.empty()) {
+        return json();
     }
-    if (value.is_string()) {
-        auto value_str = value.get<std::string>();
-        if (value_str == "reasoning_content") {
-            return COMMON_CHAT_CONTINUATION_REASONING;
-        }
-        if (value_str == "content") {
-            return COMMON_CHAT_CONTINUATION_CONTENT;
-        }
+
+    auto result = json::array();
+    for (const auto & tool : tools) {
+        result.push_back({
+            { "type",     "function" },
+            { "function",
+             {
+                  { "name", tool.name },
+                  { "description", tool.description },
+                  { "parameters", json::parse(tool.parameters) },
+              }                      },
+        });
     }
-    return COMMON_CHAT_CONTINUATION_NONE;
+    return result;
+}
+
+json common_chat_msg_diff_to_json_oaicompat(const common_chat_msg_diff & diff) {
+    json delta = json::object();
+    if (!diff.reasoning_content_delta.empty()) {
+        delta["reasoning_content"] = diff.reasoning_content_delta;
+    }
+    if (!diff.content_delta.empty()) {
+        delta["content"] = diff.content_delta;
+    }
+    if (diff.tool_call_index != std::string::npos) {
+        json tool_call;
+        tool_call["index"] = diff.tool_call_index;
+        if (!diff.tool_call_delta.id.empty()) {
+            tool_call["id"]   = diff.tool_call_delta.id;
+            tool_call["type"] = "function";
+        }
+        if (!diff.tool_call_delta.name.empty() || !diff.tool_call_delta.arguments.empty()) {
+            json function = json::object();
+            if (!diff.tool_call_delta.name.empty()) {
+                function["name"] = diff.tool_call_delta.name;
+            }
+            if (!diff.tool_call_delta.arguments.empty()) {
+                function["arguments"] = diff.tool_call_delta.arguments;
+            }
+            tool_call["function"] = function;
+        }
+        delta["tool_calls"] = json::array({ tool_call });
+    }
+    return delta;
 }
 
 bool common_chat_verify_template(const std::string & tmpl, bool use_jinja) {
@@ -619,26 +575,6 @@ bool common_chat_templates_was_explicit(const struct common_chat_templates * tmp
     return tmpls->has_explicit_template;
 }
 
-// LFM2 format detection: template uses <|tool_list_start|>[...]<|tool_list_end|> around the tool list
-// and <|tool_call_start|>[...]<|tool_call_end|> around each tool call
-static bool is_lfm2_template(const std::string & src) {
-    return src.find("<|tool_list_start|>") != std::string::npos &&
-           src.find("<|tool_list_end|>")   != std::string::npos;
-}
-
-common_chat_prompt_preset common_chat_get_asr_prompt(const common_chat_templates * chat_templates) {
-    common_chat_prompt_preset asr_preset;
-    asr_preset.system = "";
-    asr_preset.user   = "Transcribe audio to text";
-
-    if (chat_templates && chat_templates->template_default && is_lfm2_template(chat_templates->template_default->source())) {
-        asr_preset.system = "Perform ASR.";
-        asr_preset.user   = "";
-    }
-
-    return asr_preset;
-}
-
 std::string common_chat_templates_source(const struct common_chat_templates * tmpls, const std::string & variant) {
     if (!variant.empty()) {
         if (variant == "tool_use") {
@@ -886,36 +822,6 @@ std::string common_chat_template_direct_apply(
     return common_chat_template_direct_apply_impl(tmpl, inputs, std::nullopt, std::nullopt, std::nullopt);
 }
 
-static std::string common_chat_template_generation_prompt_impl(
-    const common_chat_template & tmpl,
-    const autoparser::generation_params & inputs,
-    const std::optional<json> & messages_override = std::nullopt,
-    const std::optional<json> & tools_override = std::nullopt,
-    const std::optional<json> & additional_context = std::nullopt) {
-
-    auto adjusted_messages = messages_override ? *messages_override : inputs.messages;
-
-    autoparser::generation_params params = inputs;
-    params.add_generation_prompt = false;
-    params.continue_final_message = COMMON_CHAT_CONTINUATION_NONE;
-    std::string no_gen_prompt    = common_chat_template_direct_apply_impl(tmpl, params, adjusted_messages, tools_override, additional_context);
-    params.add_generation_prompt = true;
-    std::string gen_prompt       = common_chat_template_direct_apply_impl(tmpl, params, adjusted_messages, tools_override, additional_context);
-
-    size_t prefix_len = 0;
-    size_t min_size = std::min(no_gen_prompt.size(), gen_prompt.size());
-    while (prefix_len < min_size && no_gen_prompt[prefix_len] == gen_prompt[prefix_len]) {
-        prefix_len++;
-    }
-    return gen_prompt.substr(prefix_len);
-}
-
-std::string common_chat_template_generation_prompt(
-    const common_chat_template & tmpl,
-    const autoparser::generation_params & inputs) {
-    return common_chat_template_generation_prompt_impl(tmpl, inputs, std::nullopt, std::nullopt, std::nullopt);
-}
-
 static common_chat_params common_chat_params_init_ministral_3(const common_chat_template &    tmpl,
                                                               const autoparser::generation_params & inputs) {
     common_chat_params data;
@@ -968,7 +874,6 @@ static common_chat_params common_chat_params_init_ministral_3(const common_chat_
     data.thinking_start_tag = "[THINK]";
     data.thinking_end_tag   = "[/THINK]";
     data.prompt            = common_chat_template_direct_apply_impl(tmpl, inputs, /* messages_override = */ adjusted_messages);
-    data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs, /* messages_override = */ adjusted_messages);
     data.format            = COMMON_CHAT_FORMAT_PEG_NATIVE;
     data.preserved_tokens  = {
         "[THINK]",
@@ -977,19 +882,8 @@ static common_chat_params common_chat_params_init_ministral_3(const common_chat_
         "[ARGS]",
     };
 
-    if (inputs.has_continuation()) {
-        const auto & msg = inputs.continue_msg;
-
-        data.generation_prompt = "[THINK]" + msg.reasoning_content;
-        if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
-            data.generation_prompt += "[/THINK]" + msg.render_content();
-        }
-
-        data.prompt += data.generation_prompt;
-    }
-
     auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
-        auto generation_prompt = p.eps();
+        auto generation_prompt = p.prefix(inputs.generation_prompt, "[THINK]");
         auto reasoning =
             extract_reasoning ? p.optional("[THINK]" + p.reasoning(p.until("[/THINK]")) + "[/THINK]") : p.eps();
 
@@ -1080,15 +974,6 @@ static common_chat_params common_chat_params_init_gpt_oss(const common_chat_temp
     }
 
     data.prompt            = prompt;
-    data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs, /* messages_override= */ adjusted_messages);
-    data.message_spans = common_chat_split_by_role(prompt, {
-        { "assistant", "<|start|>assistant" },
-        { "user",      "<|start|>user"      },
-        { "system",    "<|start|>developer" },
-        { "system",    "<|start|>system"    },
-        { "tool",      "<|start|>functions" },
-    });
-
     data.format            = COMMON_CHAT_FORMAT_PEG_NATIVE;
     data.supports_thinking = true;
 
@@ -1098,18 +983,6 @@ static common_chat_params common_chat_params_init_gpt_oss(const common_chat_temp
         "<|channel|>", "<|constrain|>", "<|message|>", "<|start|>", "<|end|>",
     };
 
-    // Adjust prompt for continuation
-    if (inputs.has_continuation()) {
-        const auto & msg = inputs.continue_msg;
-
-        data.generation_prompt = "<|start|>assistant<|channel|>analysis<|message|>" + msg.reasoning_content;
-        if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
-            data.generation_prompt += "<|end|><|start|>assistant<|channel|>final<|message|>" + msg.render_content();
-        }
-
-        data.prompt += data.generation_prompt;
-    }
-
     auto has_tools           = inputs.tools.is_array() && !inputs.tools.empty();
     auto has_response_format = !inputs.json_schema.is_null() && inputs.json_schema.is_object();
     auto include_grammar     = has_response_format || (has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE);
@@ -1218,21 +1091,14 @@ static common_chat_params common_chat_params_init_gemma4(const common_chat_templ
     common_chat_params data;
 
     data.prompt            = common_chat_template_direct_apply_impl(tmpl, inputs);
-    data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs);
 
     if (inputs.add_generation_prompt && string_ends_with(data.prompt, "<turn|>\n")) {
         // This may happen if the model generates content + tool_call, the
         // template does not add the model's next turn and confuses the model
         // from emitting its proper reasoning token sequence.
-        data.generation_prompt = "<|turn>model\n";
-        data.prompt += data.generation_prompt;
+        data.prompt += "<|turn>model\n";
     }
 
-    data.message_spans = common_chat_split_by_role(data.prompt, {
-        { "user",      "<|turn>user\n"  },
-        { "assistant", "<|turn>model\n" },
-    });
-
     data.format            = COMMON_CHAT_FORMAT_PEG_GEMMA4;
     data.supports_thinking  = true;
     data.thinking_start_tag = "<|channel>thought";
@@ -1246,25 +1112,13 @@ static common_chat_params common_chat_params_init_gemma4(const common_chat_templ
         "<|turn>",
     };
 
-    if (inputs.has_continuation()) {
-        const auto & msg = inputs.continue_msg;
-
-        data.generation_prompt = string_ends_with(data.prompt, "<turn|>\n") ? "<|turn>model\n" : "";
-        data.generation_prompt += "<|channel>thought\n" + msg.reasoning_content;
-        if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
-            data.generation_prompt += "<channel|>" + msg.render_content();
-        }
-
-        data.prompt += data.generation_prompt;
-    }
-
     auto has_tools           = inputs.tools.is_array() && !inputs.tools.empty();
     auto has_response_format = !inputs.json_schema.is_null() && inputs.json_schema.is_object();
     auto include_grammar     = has_response_format || (has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE);
     auto extract_reasoning   = inputs.reasoning_format != COMMON_REASONING_FORMAT_NONE;
 
     auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
-        auto start = p.rule("start", p.optional(p.literal("<|turn>model\n")));
+        auto start = p.rule("start", p.prefix(inputs.generation_prompt, "<|channel>"));
 
         if (extract_reasoning) {
             p.rule("thought", p.literal("<|channel>thought") + p.space() + p.reasoning(p.until("<channel|>")) + p.literal("<channel|>"));
@@ -1381,22 +1235,15 @@ static common_chat_params common_chat_params_init_functionary_v3_2(const common_
                                                                    const autoparser::generation_params & inputs) {
     common_chat_params data;
 
-    data.prompt            = common_chat_template_direct_apply_impl(tmpl, inputs);
-    data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs);
-    data.format            = COMMON_CHAT_FORMAT_PEG_NATIVE;
-    data.preserved_tokens  = {
+    data.prompt           = common_chat_template_direct_apply_impl(tmpl, inputs);
+    data.format           = COMMON_CHAT_FORMAT_PEG_NATIVE;
+    data.preserved_tokens = {
         ">>>all",
     };
 
     auto has_tools         = inputs.tools.is_array() && !inputs.tools.empty();
     auto include_grammar   = has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE;
 
-    if (inputs.has_continuation()) {
-        const auto & msg = inputs.continue_msg;
-        data.generation_prompt = "<|start_header_id|>assistant<|end_header_id|>\n\n>>>all\n" + msg.render_content();
-        data.prompt += data.generation_prompt;
-    }
-
     auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
         // Functionary v3.2 format:
         // - Normal content: >>>all\n{content}
@@ -1408,7 +1255,7 @@ static common_chat_params common_chat_params_init_functionary_v3_2(const common_
         // When no tools, content goes until end
         auto content_until_tool = p.literal("all\n") + p.content(p.until(">>>"));
         auto content_until_end  = p.literal("all\n") + p.content(p.rest());
-        auto generation_prompt  = p.literal("<|start_header_id|>assistant<|end_header_id|>\n\n>>>");
+        auto generation_prompt  = p.literal(inputs.generation_prompt);
 
         // If no tools or tool_choice is NONE, just parse content
         if (!has_tools || inputs.tool_choice == COMMON_CHAT_TOOL_CHOICE_NONE) {
@@ -1482,10 +1329,9 @@ static common_chat_params common_chat_params_init_kimi_k2(const common_chat_temp
                                                           const autoparser::generation_params & inputs) {
     common_chat_params data;
 
-    data.prompt            = common_chat_template_direct_apply_impl(tmpl, inputs);
-    data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs);
-    data.format            = COMMON_CHAT_FORMAT_PEG_NATIVE;
-    data.supports_thinking = true;
+    data.prompt             = common_chat_template_direct_apply_impl(tmpl, inputs);
+    data.format             = COMMON_CHAT_FORMAT_PEG_NATIVE;
+    data.supports_thinking  = true;
     data.preserved_tokens  = {
         "<|tool_calls_section_begin|>",
         "<|tool_calls_section_end|>",
@@ -1508,22 +1354,10 @@ static common_chat_params common_chat_params_init_kimi_k2(const common_chat_temp
 
     const std::string THINK_START = "<think>";
     const std::string THINK_END   = "</think>";
-    const std::string GEN_PROMPT  = "<|im_assistant|>assistant<|im_middle|>";
 
     data.thinking_start_tag = THINK_START;
     data.thinking_end_tag   = THINK_END;
 
-    if (inputs.has_continuation()) {
-        const auto & msg = inputs.continue_msg;
-
-        data.generation_prompt = GEN_PROMPT + THINK_START + msg.reasoning_content;
-        if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
-            data.generation_prompt += THINK_END + msg.render_content();
-        }
-
-        data.prompt += data.generation_prompt;
-    }
-
     auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
         // Kimi K2 Thinking format:
         // - Reasoning: <think>{reasoning}</think>
@@ -1543,7 +1377,7 @@ static common_chat_params common_chat_params_init_kimi_k2(const common_chat_temp
         auto reasoning = extract_reasoning ? p.optional(THINK_START + p.reasoning(
             p.until_one_of({ THINK_END, "<|tool_calls_section_begin|>", "<|tool_call_begin|>" })) +
             p.optional(p.literal(THINK_END))) : p.eps();
-        auto generation_prompt = p.literal(GEN_PROMPT);
+        auto generation_prompt = p.prefix(inputs.generation_prompt, THINK_START);
 
 
         // Content only parser (no tools)
@@ -1619,7 +1453,6 @@ static common_chat_params common_chat_params_init_lfm2(const common_chat_templat
     common_chat_params data;
 
     data.prompt            = common_chat_template_direct_apply_impl(tmpl, inputs);
-    data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs);
     data.format            = COMMON_CHAT_FORMAT_PEG_NATIVE;
     data.supports_thinking = true;
     data.preserved_tokens  = {
@@ -1639,24 +1472,12 @@ static common_chat_params common_chat_params_init_lfm2(const common_chat_templat
     const std::string TOOL_CALL_END   = "<|tool_call_end|>";
     const std::string THINK_START     = "<think>";
     const std::string THINK_END       = "</think>";
-    const std::string GEN_PROMPT      = "<|im_start|>assistant\n";
 
     data.thinking_start_tag = THINK_START;
     data.thinking_end_tag   = THINK_END;
 
-    if (inputs.has_continuation()) {
-        const auto & msg = inputs.continue_msg;
-
-        data.generation_prompt = GEN_PROMPT + THINK_START + msg.reasoning_content;
-        if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
-            data.generation_prompt += THINK_END + msg.render_content();
-        }
-
-        data.prompt += data.generation_prompt;
-    }
-
     auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
-        auto generation_prompt = p.literal(GEN_PROMPT);
+        auto generation_prompt = p.prefix(inputs.generation_prompt, THINK_START);
         auto end = p.end();
 
         auto reasoning = p.eps();
@@ -1711,7 +1532,6 @@ static common_chat_params common_chat_params_init_lfm2_5(const common_chat_templ
     common_chat_params data;
 
     data.prompt            = common_chat_template_direct_apply_impl(tmpl, inputs);
-    data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs);
     data.format            = COMMON_CHAT_FORMAT_PEG_NATIVE;
     data.supports_thinking = true;
     data.preserved_tokens  = {
@@ -1727,24 +1547,12 @@ static common_chat_params common_chat_params_init_lfm2_5(const common_chat_templ
 
     const std::string THINK_START     = "<think>";
     const std::string THINK_END       = "</think>";
-    const std::string GEN_PROMPT      = "<|im_start|>assistant\n";
 
     data.thinking_start_tag = THINK_START;
     data.thinking_end_tag   = THINK_END;
 
-    if (inputs.has_continuation()) {
-        const auto & msg = inputs.continue_msg;
-
-        data.generation_prompt = GEN_PROMPT + THINK_START + msg.reasoning_content;
-        if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
-            data.generation_prompt += THINK_END + msg.render_content();
-        }
-
-        data.prompt += data.generation_prompt;
-    }
-
     auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
-        auto generation_prompt = p.literal(GEN_PROMPT);
+        auto generation_prompt = p.prefix(inputs.generation_prompt, THINK_START);
         auto end = p.end();
 
         auto reasoning = p.eps();
@@ -1795,7 +1603,6 @@ static common_chat_params common_chat_params_init_gigachat_v3(
     common_chat_params data;
 
     data.prompt            = common_chat_template_direct_apply_impl(tmpl, inputs);
-    data.generation_prompt = common_chat_template_generation_prompt_impl(tmpl, inputs);
     data.format            = COMMON_CHAT_FORMAT_PEG_NATIVE;
     data.supports_thinking = false;
     data.preserved_tokens  = {
@@ -1803,12 +1610,6 @@ static common_chat_params common_chat_params_init_gigachat_v3(
         "<|role_sep|>\n",
     };
 
-    if (inputs.has_continuation()) {
-        const auto & msg = inputs.continue_msg;
-        data.generation_prompt = "assistant<|role_sep|>\n" + msg.render_content();
-        data.prompt += data.generation_prompt;
-    }
-
     auto has_tools         = inputs.tools.is_array() && !inputs.tools.empty();
     auto include_grammar   = has_tools && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE;
     const auto *tool_call_start_prefix = "<|message_sep|>\n\nfunction call<|role_sep|>\n";
@@ -1844,7 +1645,7 @@ static common_chat_params common_chat_params_init_gigachat_v3(
             ret = p.content(p.rest());
         }
 
-        return p.literal("assistant<|role_sep|>\n") + ret;
+        return p.literal(inputs.generation_prompt) + ret;
     });
 
     data.parser = parser.save();
@@ -1872,13 +1673,12 @@ static common_chat_params common_chat_params_init_deepseek_v3_2(const common_cha
                                                                  const autoparser::generation_params & inputs) {
     common_chat_params data;
 
-    data.prompt             = common_chat_template_direct_apply_impl(tmpl, inputs);
-    data.generation_prompt  = common_chat_template_generation_prompt_impl(tmpl, inputs);
-    data.format             = COMMON_CHAT_FORMAT_PEG_NATIVE;
-    data.supports_thinking  = true;
+    data.prompt            = common_chat_template_direct_apply_impl(tmpl, inputs);
+    data.format            = COMMON_CHAT_FORMAT_PEG_NATIVE;
+    data.supports_thinking = true;
     data.thinking_start_tag = "<think>";
     data.thinking_end_tag   = "</think>";
-    data.preserved_tokens   = {
+    data.preserved_tokens  = {
         "｜DSML｜",
         "<think>",
         "</think>",
@@ -1898,21 +1698,9 @@ static common_chat_params common_chat_params_init_deepseek_v3_2(const common_cha
     const std::string INVOKE_END   = "</" + DSML + "invoke>";
     const std::string PARAM_START  = "<" + DSML + "parameter";
     const std::string PARAM_END    = "</" + DSML + "parameter>";
-    const std::string GEN_PROMPT   = "<｜Assistant｜>";
-
-    if (inputs.has_continuation()) {
-        const auto & msg = inputs.continue_msg;
-
-        data.generation_prompt = GEN_PROMPT + THINK_START + msg.reasoning_content;
-        if (inputs.continue_final_message == COMMON_CHAT_CONTINUATION_CONTENT) {
-            data.generation_prompt += THINK_END + msg.render_content();
-        }
-
-        data.prompt += data.generation_prompt;
-    }
 
     auto parser = build_chat_peg_parser([&](common_chat_peg_builder & p) {
-        auto generation_prompt = p.literal(GEN_PROMPT);
+        auto generation_prompt = p.prefix(inputs.generation_prompt, THINK_START);
         auto end = p.end();
 
         auto reasoning = p.eps();
@@ -2296,7 +2084,10 @@ std::optional<common_chat_params> common_chat_try_specialized_template(
         return common_chat_params_init_kimi_k2(tmpl, params);
     }
 
-    if (is_lfm2_template(src)) {
+    // LFM2 format detection: template uses <|tool_list_start|>[...]<|tool_list_end|> around the tool list
+    // and <|tool_call_start|>[...]<|tool_call_end|> around each tool call
+    if (src.find("<|tool_list_start|>") != std::string::npos &&
+        src.find("<|tool_list_end|>") != std::string::npos) {
         LOG_DBG("Using specialized template: LFM2\n");
         return common_chat_params_init_lfm2(tmpl, params);
     }
@@ -2345,38 +2136,16 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
     params.tools = common_chat_tools_to_json_oaicompat(inputs.tools);
     const auto & tmpl =
         params.tools.is_array() && tmpls->template_tool_use ? *tmpls->template_tool_use : *tmpls->template_default;
-    const auto & src             = tmpl.source();
-    const auto & caps            = tmpl.original_caps();
-    params.messages              = render_message_to_json(inputs.messages, tmpl.original_caps());
-    params.tool_choice           = inputs.tool_choice;
-    params.reasoning_format      = inputs.reasoning_format;
-    params.enable_thinking       = inputs.enable_thinking;
-    params.grammar               = inputs.grammar;
-    params.now                   = inputs.now;
-    params.add_generation_prompt = inputs.add_generation_prompt;
-    params.add_bos               = tmpls->add_bos;
-    params.add_eos               = tmpls->add_eos;
-
-    params.continue_final_message = inputs.continue_final_message;
-    if (params.continue_final_message != COMMON_CHAT_CONTINUATION_NONE) {
-        params.add_generation_prompt = false;
-
-        if (!inputs.messages.empty()) {
-            // Render messages[:-1] and store continuation message separately
-            params.continue_msg = inputs.messages.back();
-            params.messages.erase(params.messages.size() - 1);
-        }
-
-        if (params.continue_final_message == COMMON_CHAT_CONTINUATION_AUTO && !inputs.messages.empty()) {
-            // Resolve based on message content
-            params.continue_final_message = COMMON_CHAT_CONTINUATION_CONTENT;
-            if (!params.continue_msg.reasoning_content.empty() &&
-                params.continue_msg.content.empty() &&
-                params.continue_msg.content_parts.empty()) {
-                params.continue_final_message = COMMON_CHAT_CONTINUATION_REASONING;
-            }
-        }
-    }
+    const auto & src        = tmpl.source();
+    const auto & caps       = tmpl.original_caps();
+    params.messages         = render_message_to_json(inputs.messages, tmpl.original_caps());
+    params.tool_choice      = inputs.tool_choice;
+    params.reasoning_format = inputs.reasoning_format;
+    params.enable_thinking  = inputs.enable_thinking;
+    params.grammar          = inputs.grammar;
+    params.now              = inputs.now;
+    params.add_bos          = tmpls->add_bos;
+    params.add_eos          = tmpls->add_eos;
 
     if (src.find("<|channel|>") == std::string::npos) {
         // map developer to system for all models except for GPT-OSS
@@ -2398,6 +2167,15 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
         workaround::func_args_not_string(params.messages);
     }
 
+    params.add_generation_prompt = false;
+    std::string no_gen_prompt    = common_chat_template_direct_apply_impl(tmpl, params);
+    params.add_generation_prompt = true;
+    std::string gen_prompt       = common_chat_template_direct_apply_impl(tmpl, params);
+    auto        diff             = calculate_diff_split(no_gen_prompt, gen_prompt);
+    params.generation_prompt     = diff.right + diff.suffix;
+
+    params.add_generation_prompt = inputs.add_generation_prompt;
+
     params.extra_context = common_chat_extra_context();
     for (auto el : inputs.chat_template_kwargs) {
         params.extra_context[el.first] = json::parse(el.second);
@@ -2427,16 +2205,17 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
         auto params_copy               = params;
         params_copy.reasoning_format   = COMMON_REASONING_FORMAT_NONE;
         data.prompt                    = common_chat_template_direct_apply_impl(tmpl, params_copy);
-        data.generation_prompt         = common_chat_template_generation_prompt_impl(tmpl, params);
         data.format                    = COMMON_CHAT_FORMAT_PEG_NATIVE;
-        auto parser                    = build_chat_peg_parser([&data](common_chat_peg_builder &p) {
-            return p.literal(data.generation_prompt) << p.content(p.rest());
+        data.generation_prompt         = params.generation_prompt;
+        auto parser                    = build_chat_peg_parser([&params](common_chat_peg_builder &p) {
+            return p.prefix(params.generation_prompt) << p.content(p.rest());
         });
         data.parser                    = parser.save();
         return data;
     }
 
     if (auto result = common_chat_try_specialized_template(tmpl, src, params)) {
+        result->generation_prompt = params.generation_prompt;
         return *result;
     }
 
@@ -2445,24 +2224,12 @@ static common_chat_params common_chat_templates_apply_jinja(const struct common_
         struct autoparser::autoparser autoparser;
         autoparser.analyze_template(tmpl);
         auto auto_params = autoparser::peg_generator::generate_parser(tmpl, params, autoparser);
-
-        std::vector<common_chat_msg_delimiter> delimiters;
-        if (!autoparser.assistant_start.empty()) {
-            delimiters.push_back({ "assistant", autoparser.assistant_start });
-        }
-        if (!autoparser.user_start.empty()) {
-            delimiters.push_back({ "user", autoparser.user_start });
-        }
-
-        if (!delimiters.empty()) {
-            auto_params.message_spans = common_chat_split_by_role(auto_params.prompt, delimiters);
-        }
-
         auto_params.supports_thinking = autoparser.reasoning.mode != autoparser::reasoning_mode::NONE;
         if (auto_params.supports_thinking) {
-            auto_params.thinking_start_tag = trim_whitespace(autoparser.reasoning.start);
-            auto_params.thinking_end_tag   = trim_whitespace(autoparser.reasoning.end);
+            auto_params.thinking_start_tag = autoparser.reasoning.start;
+            auto_params.thinking_end_tag   = autoparser.reasoning.end;
         }
+        auto_params.generation_prompt = params.generation_prompt;
         common_peg_arena arena;
         arena.load(auto_params.parser);
         LOG_DBG("%s: generated parser:\n%s\n\nparser generation prompt: %s\n", __func__, arena.dump(arena.root()).c_str(), auto_params.generation_prompt.c_str());
@@ -2629,3 +2396,4 @@ std::map<std::string, bool> common_chat_templates_get_caps(const common_chat_tem
     GGML_ASSERT(chat_templates->template_default != nullptr);
     return chat_templates->template_default->caps.to_map();
 }
+

common/chat.h

@@ -89,22 +89,11 @@ struct common_chat_msg {
 
     nlohmann::ordered_json to_json_oaicompat(bool concat_typed_text = false) const;
 
-    std::string render_content(const std::string & delimiter = "\n\n") const;
-
     bool empty() const {
         return content.empty() && content_parts.empty() && tool_calls.empty() && reasoning_content.empty() &&
                tool_name.empty() && tool_call_id.empty();
     }
 
-    bool contains_media() const {
-        for (const auto & part : content_parts) {
-            if (part.type == "media_marker") {
-                return true;
-            }
-        }
-        return false;
-    }
-
     void set_tool_call_ids(std::vector<std::string> &           ids_cache,
                            const std::function<std::string()> & gen_tool_call_id) {
         for (auto i = 0u; i < tool_calls.size(); i++) {
@@ -143,17 +132,6 @@ struct common_chat_msg_diff {
     }
 };
 
-struct common_chat_msg_span {
-    std::string role;
-    std::size_t pos = 0;
-    std::size_t len = 0;
-};
-
-struct common_chat_msg_delimiter {
-    std::string role;
-    std::string delimiter;
-};
-
 struct common_chat_tool {
     std::string name;
     std::string description;
@@ -177,22 +155,12 @@ enum common_chat_format {
     COMMON_CHAT_FORMAT_COUNT,  // Not a format, just the # formats
 };
 
-
-// Continuation method provided via `continue_final_message`
-enum common_chat_continuation {
-    COMMON_CHAT_CONTINUATION_NONE,
-    COMMON_CHAT_CONTINUATION_AUTO,
-    COMMON_CHAT_CONTINUATION_REASONING,
-    COMMON_CHAT_CONTINUATION_CONTENT,
-};
-
 struct common_chat_templates_inputs {
     std::vector<common_chat_msg>          messages;
     std::string                           grammar;
     std::string                           json_schema;
-    bool                                  add_generation_prompt  = true;
-    common_chat_continuation              continue_final_message = COMMON_CHAT_CONTINUATION_NONE;
-    bool                                  use_jinja              = true;
+    bool                                  add_generation_prompt = true;
+    bool                                  use_jinja             = true;
     // Parameters below only supported when use_jinja is true
     std::vector<common_chat_tool>         tools;
     common_chat_tool_choice               tool_choice         = COMMON_CHAT_TOOL_CHOICE_AUTO;
@@ -219,7 +187,6 @@ struct common_chat_params {
     std::vector<std::string>            preserved_tokens;
     std::vector<std::string>            additional_stops;
     std::string                         parser;
-    std::vector<common_chat_msg_span>   message_spans;
 };
 
 // per-message parsing syntax
@@ -231,8 +198,6 @@ struct common_chat_parser_params {
     bool                    reasoning_in_content = false;
     std::string             generation_prompt;
     bool                    parse_tool_calls     = true;
-    bool                    is_continuation      = false;
-    bool                    echo                 = false;  // Include assistant prefilled msg in output
     bool                    debug                = false;  // Enable debug output for PEG parser
     common_peg_arena        parser               = {};
     common_chat_parser_params() = default;
@@ -291,15 +256,14 @@ bool common_chat_templates_support_enable_thinking(const common_chat_templates *
 // Parses a JSON array of messages in OpenAI's chat completion API format.
 std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const nlohmann::ordered_json & messages);
 
-std::vector<common_chat_tool> common_chat_tools_parse_oaicompat(const nlohmann::ordered_json & tools);
-
-common_chat_continuation common_chat_continuation_parse(const nlohmann::ordered_json & value);
-
 // DEPRECATED: only used in tests
 nlohmann::ordered_json common_chat_msgs_to_json_oaicompat(const std::vector<common_chat_msg> & msgs, bool concat_typed_text = false);
 
+std::vector<common_chat_tool> common_chat_tools_parse_oaicompat(const nlohmann::ordered_json & tools);
 nlohmann::ordered_json common_chat_tools_to_json_oaicompat(const std::vector<common_chat_tool> & tools);
 
+nlohmann::ordered_json common_chat_msg_diff_to_json_oaicompat(const common_chat_msg_diff & diff);
+
 // get template caps, useful for reporting to server /props endpoint
 std::map<std::string, bool> common_chat_templates_get_caps(const common_chat_templates * chat_templates);
 
@@ -307,23 +271,7 @@ std::string common_chat_template_direct_apply(
     const common_chat_template & tmpl,
     const autoparser::generation_params & inputs);
 
-std::string common_chat_template_generation_prompt(
-    const common_chat_template &          tmpl,
-    const autoparser::generation_params & inputs);
-
 std::optional<common_chat_params> common_chat_try_specialized_template(
         const common_chat_template &          tmpl,
         const std::string &                   src,
         autoparser::generation_params & params);
-
-
-// specialized per-task preset
-struct common_chat_prompt_preset {
-    std::string system;
-    std::string user;
-};
-
-common_chat_prompt_preset common_chat_get_asr_prompt(const common_chat_templates * chat_templates);
-
-std::vector<common_chat_msg_span> common_chat_split_by_role(const std::string & prompt, const std::vector<common_chat_msg_delimiter> & delims);
-

common/common.cpp

@@ -3,11 +3,9 @@
 
 #include "build-info.h"
 #include "common.h"
-#include "fit.h"
 #include "log.h"
 #include "llama.h"
 #include "sampling.h"
-#include "speculative.h"
 #include "unicode.h"
 
 #include <algorithm>
@@ -71,7 +69,7 @@ common_time_meas::~common_time_meas() {
 // CPU utils
 //
 
-int32_t common_cpu_get_num_physical_cores() {
+int32_t cpu_get_num_physical_cores() {
 #ifdef __linux__
     // enumerate the set of thread siblings, num entries is num cores
     std::unordered_set<std::string> siblings;
@@ -186,11 +184,11 @@ static int cpu_count_math_cpus(int n_cpu) {
 /**
  * Returns number of CPUs on system that are useful for math.
  */
-int32_t common_cpu_get_num_math() {
+int32_t cpu_get_num_math() {
 #if defined(__x86_64__) && defined(__linux__) && !defined(__ANDROID__)
     int n_cpu = sysconf(_SC_NPROCESSORS_ONLN);
     if (n_cpu < 1) {
-        return common_cpu_get_num_physical_cores();
+        return cpu_get_num_physical_cores();
     }
     if (is_hybrid_cpu()) {
         cpu_set_t affinity;
@@ -203,7 +201,7 @@ int32_t common_cpu_get_num_math() {
         }
     }
 #endif
-    return common_cpu_get_num_physical_cores();
+    return cpu_get_num_physical_cores();
 }
 
 // Helper for setting process priority
@@ -264,7 +262,7 @@ bool set_process_priority(enum ggml_sched_priority prio) {
 //
 
 
-void postprocess_cpu_params(common_cpu_params & cpuparams, const common_cpu_params * role_model) {
+void postprocess_cpu_params(cpu_params& cpuparams, const cpu_params* role_model) {
     int32_t n_set = 0;
 
     if (cpuparams.n_threads < 0) {
@@ -272,7 +270,7 @@ void postprocess_cpu_params(common_cpu_params & cpuparams, const common_cpu_para
         if (role_model != nullptr) {
             cpuparams = *role_model;
         } else {
-            cpuparams.n_threads = common_cpu_get_num_math();
+            cpuparams.n_threads = cpu_get_num_math();
         }
     }
 
@@ -367,33 +365,15 @@ void common_init() {
     SetConsoleCP(CP_UTF8);
 #endif
 
-    common_log_set_prefix(common_log_main(), true);
-    common_log_set_timestamps(common_log_main(), true);
-
     llama_log_set(common_log_default_callback, NULL);
-}
 
-void common_params_print_info(const common_params & params, bool print_devices) {
 #ifdef NDEBUG
     const char * build_type = "";
 #else
     const char * build_type = " (debug)";
 #endif
-    LOG_TRC("%s: build %d (%s) with %s for %s%s\n", __func__, llama_build_number(), llama_commit(), llama_compiler(), llama_build_target(), build_type);
 
-    LOG_INF("log_info: verbosity = %d (adjust with the `-lv N` CLI arg)\n", common_log_get_verbosity_thold());
-
-    // device enumeration creates a primary context on CUDA backends, skip it when the caller does not own any device
-    if (print_devices) {
-        LOG_INF("device_info:\n");
-        for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
-            auto * dev = ggml_backend_dev_get(i);
-            size_t free, total;
-            ggml_backend_dev_memory(dev, &free, &total);
-            LOG_INF("  - %-8s: %s (%zu MiB, %zu MiB free)\n", ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), total / 1024 / 1024, free / 1024 / 1024);
-        }
-    }
-    LOG_INF("%s\n", common_params_get_system_info(params).c_str());
+    LOG_DBG("build: %d (%s) with %s for %s%s\n", llama_build_number(), llama_commit(), llama_compiler(), llama_build_target(), build_type);
 }
 
 std::string common_params_get_system_info(const common_params & params) {
@@ -445,27 +425,6 @@ std::string string_strip(const std::string & str) {
     return str.substr(start, end - start);
 }
 
-std::string string_lcs(std::string_view a, std::string_view b) {
-    if (a.empty() || b.empty()) return {};
-
-    std::vector<std::vector<size_t>> dp(a.size() + 1, std::vector<size_t>(b.size() + 1, 0));
-    size_t best_len = 0;
-    size_t best_end_a = 0;
-
-    for (size_t i = 1; i <= a.size(); ++i) {
-        for (size_t j = 1; j <= b.size(); ++j) {
-            if (a[i - 1] == b[j - 1]) {
-                dp[i][j] = dp[i - 1][j - 1] + 1;
-                if (dp[i][j] > best_len) {
-                    best_len = dp[i][j];
-                    best_end_a = i;
-                }
-            }
-        }
-    }
-    return std::string(a.substr(best_end_a - best_len, best_len));
-}
-
 std::string string_get_sortable_timestamp() {
     using clock = std::chrono::system_clock;
 
@@ -1181,20 +1140,19 @@ struct common_init_result::impl {
     std::vector<llama_sampler_seq_config> samplers_seq_config;
 };
 
-common_init_result::common_init_result(common_params & params, bool model_only) :
+common_init_result::common_init_result(common_params & params) :
     pimpl(new impl{}) {
     auto mparams = common_model_params_to_llama(params);
     auto cparams = common_context_params_to_llama(params);
 
     if (params.fit_params) {
-        LOG_INF("%s: fitting params to device memory ...\n", __func__);
-        LOG_INF("%s: (for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on)\n", __func__);
-        common_fit_params(params.model.path.c_str(), &mparams, &cparams,
+        LOG_INF("%s: fitting params to device memory, for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on\n", __func__);
+        llama_params_fit(params.model.path.c_str(), &mparams, &cparams,
             params.tensor_split,
             params.tensor_buft_overrides.data(),
             params.fit_params_target.data(),
             params.fit_params_min_ctx,
-            params.verbosity >= LOG_LEVEL_DEBUG ? GGML_LOG_LEVEL_DEBUG : GGML_LOG_LEVEL_ERROR);
+            params.verbosity >= 4 ? GGML_LOG_LEVEL_DEBUG : GGML_LOG_LEVEL_ERROR);
     }
 
     llama_model * model = llama_model_load_from_file(params.model.path.c_str(), mparams);
@@ -1204,10 +1162,6 @@ common_init_result::common_init_result(common_params & params, bool model_only)
 
     pimpl->model.reset(model);
 
-    if (model_only) {
-        return;
-    }
-
     const llama_vocab * vocab = llama_model_get_vocab(model);
 
     // load and optionally apply lora adapters
@@ -1241,7 +1195,7 @@ common_init_result::common_init_result(common_params & params, bool model_only)
     // initialize once
     for (llama_token i = 0; i < llama_vocab_n_tokens(vocab); i++) {
         if (llama_vocab_is_eog(vocab, i)) {
-            LOG_TRC("%s: added %s logit bias = %f\n", __func__, common_token_to_piece(vocab, i).c_str(), -INFINITY);
+            LOG_INF("%s: added %s logit bias = %f\n", __func__, common_token_to_piece(vocab, i).c_str(), -INFINITY);
             params.sampling.logit_bias_eog.push_back({i, -INFINITY});
         }
     }
@@ -1254,12 +1208,12 @@ common_init_result::common_init_result(common_params & params, bool model_only)
     }
 
     //if (params.sampling.penalty_last_n == -1) {
-    //    LOG_TRC("%s: setting penalty_last_n to ctx_size = %d\n", __func__, llama_n_ctx(lctx));
+    //    LOG_INF("%s: setting penalty_last_n to ctx_size = %d\n", __func__, llama_n_ctx(lctx));
     //    params.sampling.penalty_last_n = llama_n_ctx(lctx);
     //}
 
     //if (params.sampling.dry_penalty_last_n == -1) {
-    //    LOG_TRC("%s: setting dry_penalty_last_n to ctx_size = %d\n", __func__, llama_n_ctx(lctx));
+    //    LOG_INF("%s: setting dry_penalty_last_n to ctx_size = %d\n", __func__, llama_n_ctx(lctx));
     //    params.sampling.dry_penalty_last_n = llama_n_ctx(lctx);
     //}
 
@@ -1311,8 +1265,8 @@ std::vector<llama_adapter_lora_ptr> & common_init_result::lora() {
     return pimpl->lora;
 }
 
-common_init_result_ptr common_init_from_params(common_params & params, bool model_only) {
-    common_init_result_ptr res(new common_init_result(params, model_only));
+common_init_result_ptr common_init_from_params(common_params & params) {
+    common_init_result_ptr res(new common_init_result(params));
 
     llama_model * model = res->model();
     if (model == NULL) {
@@ -1320,10 +1274,6 @@ common_init_result_ptr common_init_from_params(common_params & params, bool mode
         return res;
     }
 
-    if (model_only) {
-        return res;
-    }
-
     llama_context * lctx = res->context();
     if (lctx == NULL) {
         LOG_ERR("%s: failed to create context with model '%s'\n", __func__, params.model.path.c_str());
@@ -1387,7 +1337,7 @@ common_init_result_ptr common_init_from_params(common_params & params, bool mode
     }
 
     if (params.warmup) {
-        LOG_INF("%s: warming up the model with an empty run - please wait ... (--no-warmup to disable)\n", __func__);
+        LOG_WRN("%s: warming up the model with an empty run - please wait ... (--no-warmup to disable)\n", __func__);
 
         llama_set_warmup(lctx, true);
 
@@ -1469,15 +1419,9 @@ common_context_seq_rm_type common_context_can_seq_rm(llama_context * ctx) {
         goto done;
     }
 
-    if (llama_n_rs_seq(ctx) > 0) {
-        LOG_INF("%s: the context supports bounded partial sequence removal\n", __func__);
-        res = COMMON_CONTEXT_SEQ_RM_TYPE_RS;
-        goto done;
-    }
-
     // try to remove the last tokens
     if (!llama_memory_seq_rm(mem, 0, 1, -1)) {
-        LOG_TRC("%s: the context does not support partial sequence removal\n", __func__);
+        LOG_WRN("%s: the target context does not support partial sequence removal\n", __func__);
         res = COMMON_CONTEXT_SEQ_RM_TYPE_FULL;
         goto done;
     }
@@ -1489,23 +1433,6 @@ done:
     return res;
 }
 
-void common_context_seq_rm(llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1) {
-    auto * mem = llama_get_memory(ctx);
-    if (!llama_memory_seq_rm(mem, seq_id, p0, p1)) {
-        GGML_ABORT("%s", string_format("failed to remove sequence %d with p0=%d, p1=%d\n", seq_id, p0, p1).c_str());
-    }
-}
-
-void common_context_seq_cp(llama_context * ctx, llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1) {
-    auto * mem = llama_get_memory(ctx);
-    llama_memory_seq_cp(mem, seq_id_src, seq_id_dst, p0, p1);
-}
-
-void common_context_seq_add(llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos delta) {
-    auto * mem = llama_get_memory(ctx);
-    llama_memory_seq_add(mem, seq_id, p0, p1, delta);
-}
-
 void common_set_adapter_lora(struct llama_context * ctx, std::vector<common_adapter_lora_info> & lora) {
     std::vector<llama_adapter_lora *> loras;
     std::vector<float> scales;
@@ -1562,7 +1489,6 @@ struct llama_context_params common_context_params_to_llama(const common_params &
 
     cparams.n_ctx             = params.n_ctx;
     cparams.n_seq_max         = params.n_parallel;
-    cparams.n_rs_seq          = params.speculative.need_n_rs_seq();
     cparams.n_batch           = params.n_batch;
     cparams.n_ubatch          = params.n_ubatch;
     cparams.n_threads         = params.cpuparams.n_threads;
@@ -1594,7 +1520,7 @@ struct llama_context_params common_context_params_to_llama(const common_params &
     return cparams;
 }
 
-struct ggml_threadpool_params ggml_threadpool_params_from_cpu_params(const common_cpu_params & params) {
+struct ggml_threadpool_params ggml_threadpool_params_from_cpu_params(const cpu_params & params) {
     struct ggml_threadpool_params tpp;
 
     ggml_threadpool_params_init(&tpp, params.n_threads); // setup the defaults
@@ -2033,110 +1959,3 @@ bool common_prompt_batch_decode(
 
     return true;
 }
-
-size_t common_prompt_checkpoint::size() const {
-    return data_tgt.size() + data_dft.size();
-}
-
-bool common_prompt_checkpoint::empty() const {
-    return data_tgt.empty();
-}
-
-void common_prompt_checkpoint::clear() {
-    n_tokens = 0;
-
-    pos_min = 0;
-    pos_max = 0;
-
-    data_tgt.clear();
-    data_dft.clear();
-}
-
-void common_prompt_checkpoint::update_pos(
-        int64_t n_tokens,
-        llama_pos pos_min,
-        llama_pos pos_max) {
-    this->n_tokens = n_tokens;
-    this->pos_min  = pos_min;
-    this->pos_max  = pos_max;
-}
-
-void common_prompt_checkpoint::update_tgt(
-        llama_context * ctx,
-        llama_seq_id seq_id,
-        llama_state_seq_flags flags) {
-    if (ctx == nullptr) {
-        return;
-    }
-
-    const size_t ckpt_size = llama_state_seq_get_size_ext(ctx, seq_id, flags);
-
-    data_tgt.resize(ckpt_size);
-
-    const size_t n = llama_state_seq_get_data_ext(ctx, data_tgt.data(), ckpt_size, seq_id, flags);
-    if (n != ckpt_size) {
-        GGML_ABORT("checkpoint size mismatch: expected %zu, got %zu\n", ckpt_size, n);
-    }
-}
-
-void common_prompt_checkpoint::update_dft(
-        llama_context * ctx,
-        llama_seq_id seq_id,
-        llama_state_seq_flags flags) {
-    if (ctx == nullptr) {
-        return;
-    }
-
-    const size_t ckpt_size = llama_state_seq_get_size_ext(ctx, seq_id, flags);
-
-    data_dft.resize(ckpt_size);
-
-    const size_t n = llama_state_seq_get_data_ext(ctx, data_dft.data(), ckpt_size, seq_id, flags);
-    if (n != ckpt_size) {
-        GGML_ABORT("checkpoint size mismatch: expected %zu, got %zu\n", ckpt_size, n);
-    }
-}
-
-void common_prompt_checkpoint::load_tgt(
-        llama_context * ctx,
-        llama_seq_id seq_id,
-        llama_state_seq_flags flags) const {
-    if (ctx == nullptr) {
-        return;
-    }
-
-    if (data_tgt.empty()) {
-        return;
-    }
-
-    const size_t n = llama_state_seq_set_data_ext(ctx, data_tgt.data(), data_tgt.size(), seq_id, flags);
-    if (n != data_tgt.size()) {
-        GGML_ABORT("checkpoint size mismatch: expected %zu, got %zu\n", data_tgt.size(), n);
-    }
-}
-
-void common_prompt_checkpoint::load_dft(
-        llama_context * ctx,
-        llama_seq_id seq_id,
-        llama_state_seq_flags flags) const {
-    if (ctx == nullptr) {
-        return;
-    }
-
-    if (data_dft.empty()) {
-        return;
-    }
-
-    const size_t n = llama_state_seq_set_data_ext(ctx, data_dft.data(), data_dft.size(), seq_id, flags);
-    if (n != data_dft.size()) {
-        GGML_ABORT("checkpoint size mismatch: expected %zu, got %zu\n", data_dft.size(), n);
-    }
-}
-
-void common_prompt_checkpoint::clear_tgt() {
-    data_tgt.clear();
-}
-
-void common_prompt_checkpoint::clear_dft() {
-    data_dft.clear();
-}

common/common.h

@@ -13,7 +13,6 @@
 #include <string_view>
 #include <vector>
 #include <map>
-#include <algorithm>
 
 #if defined(_WIN32) && !defined(_WIN32_WINNT)
 #define _WIN32_WINNT 0x0A00
@@ -55,7 +54,7 @@ struct common_control_vector_load_info;
 // CPU utils
 //
 
-struct common_cpu_params {
+struct cpu_params {
     int      n_threads                   = -1;
     bool     cpumask[GGML_MAX_N_THREADS] = {false}; // CPU affinity mask.
     bool     mask_valid                  = false;   // Default: any CPU
@@ -64,8 +63,8 @@ struct common_cpu_params {
     uint32_t poll                        = 50;      // Polling (busywait) level (0 - no polling, 100 - mostly polling)
 };
 
-int32_t common_cpu_get_num_physical_cores();
-int32_t common_cpu_get_num_math();
+int32_t cpu_get_num_physical_cores();
+int32_t cpu_get_num_math();
 
 //
 // Common params
@@ -158,10 +157,9 @@ enum common_params_sampling_config : uint64_t {
 
 enum common_speculative_type {
     COMMON_SPECULATIVE_TYPE_NONE,          // no speculative decoding
-    COMMON_SPECULATIVE_TYPE_DRAFT_SIMPLE,  // standalone draft model speculative decoding
-    COMMON_SPECULATIVE_TYPE_DRAFT_EAGLE3,  // Eagle3 speculative decoding
-    COMMON_SPECULATIVE_TYPE_DRAFT_MTP,     // Multi-token prediction
-    COMMON_SPECULATIVE_TYPE_NGRAM_SIMPLE,  // simple self-speculative decoding based on n-grams
+    COMMON_SPECULATIVE_TYPE_DRAFT,         // draft model
+    COMMON_SPECULATIVE_TYPE_EAGLE3,        // eagle draft model
+    COMMON_SPECULATIVE_TYPE_NGRAM_SIMPLE,  // simple self-speculative decoding
     COMMON_SPECULATIVE_TYPE_NGRAM_MAP_K,   // self-speculative decoding with n-gram keys only
     COMMON_SPECULATIVE_TYPE_NGRAM_MAP_K4V, // self-speculative decoding with n-gram keys and 4 m-gram values
     COMMON_SPECULATIVE_TYPE_NGRAM_MOD,
@@ -276,7 +274,6 @@ struct common_params_sampling {
     std::vector<llama_token> reasoning_budget_start;           // start tag token sequence
     std::vector<llama_token> reasoning_budget_end;             // end tag token sequence
     std::vector<llama_token> reasoning_budget_forced;          // forced sequence (message + end tag)
-    std::string              reasoning_budget_message;         // message injected before end tag when budget exhausted
 
     bool backend_sampling = false;
 
@@ -297,84 +294,62 @@ struct common_params_model {
     std::string name        = ""; // in format <user>/<model>[:<tag>] (tag is optional)     // NOLINT
 };
 
-// draft-model-based speculative decoding parameters
-struct common_params_speculative_draft {
-    int32_t n_max = 3; // maximum number of tokens to draft during speculative decoding
-    int32_t n_min = 0; // minimum number of draft tokens to use for speculative decoding
+struct common_ngram_mod;
 
-    float p_split = 0.1f; // speculative decoding split probability
-    float p_min   = 0.0f; // minimum speculative decoding probability (greedy)
+struct common_params_speculative {
+    common_speculative_type type = COMMON_SPECULATIVE_TYPE_NONE; // type of speculative decoding
 
-    bool backend_sampling = true; // offload draft sampling to the backend (default: on)
+    // general-purpose speculative decoding parameters
 
-    common_params_model mparams;
+    int32_t n_max   = 16; // maximum number of tokens to draft during speculative decoding
+    int32_t n_min   = 0;  // minimum number of draft tokens to use for speculative decoding
+    float   p_split = 0.1f; // speculative decoding split probability
+    float   p_min   = 0.75f; // minimum speculative decoding probability (greedy)
 
-    llama_context * ctx_tgt = nullptr;
-    llama_context * ctx_dft = nullptr;
+    // ngram-based speculative decoding
 
+    uint16_t ngram_size_n   = 12; // ngram size for lookup
+    uint16_t ngram_size_m   = 48; // mgram size for speculative tokens
+    uint16_t ngram_min_hits = 1; // minimum hits at ngram/mgram lookup for mgram to be proposed
+
+    std::shared_ptr<common_ngram_mod> ngram_mod;
+
+    std::string lookup_cache_static;  // path of static ngram cache file for lookup decoding           // NOLINT
+    std::string lookup_cache_dynamic; // path of dynamic ngram cache file for lookup decoding          // NOLINT
+
+    // draft-model speculative decoding
+
+    struct common_params_model mparams_dft;
+
+    llama_model * model_dft = nullptr; // a llama_model that can be shared by multiple speculative contexts
+
+    llama_context_params cparams_dft; // these are the parameters for the draft llama_context
+
+    int32_t n_ctx        = 0;  // draft context size
     int32_t n_gpu_layers = -1; // number of layers to store in VRAM for the draft model (-1 - use default)
 
     ggml_type cache_type_k = GGML_TYPE_F16; // KV cache data type for the K
     ggml_type cache_type_v = GGML_TYPE_F16; // KV cache data type for the V
 
-    common_cpu_params cpuparams;
-    common_cpu_params cpuparams_batch;
+    struct cpu_params cpuparams;
+    struct cpu_params cpuparams_batch;
 
     std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
 
+    std::vector<std::pair<std::string, std::string>> replacements; // main to speculative model replacements
     std::vector<llama_model_tensor_buft_override> tensor_buft_overrides;
-};
-
-struct common_params_speculative_ngram_mod {
-    int32_t n_match = 24;
-
-    int32_t n_max = 64;
-    int32_t n_min = 48;
-};
-
-struct common_params_speculative_ngram_map {
-    uint16_t size_n   = 12; // ngram size for lookup
-    uint16_t size_m   = 48; // mgram size for speculative tokens
-    uint16_t min_hits = 1;  // minimum hits at ngram/mgram lookup for mgram to be proposed
-};
-
-struct common_params_speculative_ngram_cache {
-    std::string lookup_cache_static;  // path of static ngram cache file for lookup decoding
-    std::string lookup_cache_dynamic; // path of dynamic ngram cache file for lookup decoding
-};
-
-struct common_params_speculative {
-    std::vector<enum common_speculative_type> types = { COMMON_SPECULATIVE_TYPE_NONE };
-
-    // used by Simple, MTP, Eagle3, etc. - all methods that require some kind of draft model
-    common_params_speculative_draft draft;
-
-    common_params_speculative_ngram_mod ngram_mod;
-    common_params_speculative_ngram_map ngram_simple;
-    common_params_speculative_ngram_map ngram_map_k;
-    common_params_speculative_ngram_map ngram_map_k4v;
-
-    common_params_speculative_ngram_cache ngram_cache;
 
     bool has_dft() const {
-        return !draft.mparams.path.empty() || !draft.mparams.hf_repo.empty();
-    }
-
-    uint32_t need_n_rs_seq() const {
-        bool needs_rs_seq = std::any_of(types.begin(), types.end(), [&](auto t) {
-            return t == COMMON_SPECULATIVE_TYPE_DRAFT_MTP;
-        });
-
-        return needs_rs_seq ? draft.n_max : 0u;
+        return !mparams_dft.path.empty() || !mparams_dft.hf_repo.empty();
     }
 };
 
 struct common_params_vocoder {
     struct common_params_model model;
 
-    std::string speaker_file; // speaker file path
+    std::string speaker_file = ""; // speaker file path                                      // NOLINT
 
-    bool use_guide_tokens = false; // enable guide tokens to improve TTS accuracy
+    bool use_guide_tokens = false; // enable guide tokens to improve TTS accuracy            // NOLINT
 };
 
 struct common_params_diffusion {
@@ -445,20 +420,19 @@ struct common_params {
     // offload params
     std::vector<ggml_backend_dev_t> devices; // devices to use for offloading
 
-    int32_t n_gpu_layers       = -1;    // number of layers to store in VRAM, -1 is auto, <= -2 is all
-    int32_t main_gpu           = 0;     // the GPU that is used for scratch and small tensors
-    float   tensor_split[128]  = {0};   // how split tensors should be distributed across GPUs
-    bool    fit_params         = true;  // whether to fit unset model/context parameters to free device memory
-    bool    fit_params_print   = false; // print the estimated required memory to run the model
-    int32_t fit_params_min_ctx = 4096;  // minimum context size to set when trying to reduce memory use
+    int32_t n_gpu_layers       = -1;   // number of layers to store in VRAM, -1 is auto, <= -2 is all
+    int32_t main_gpu           = 0;    // the GPU that is used for scratch and small tensors
+    float   tensor_split[128]  = {0};  // how split tensors should be distributed across GPUs
+    bool    fit_params         = true; // whether to fit unset model/context parameters to free device memory
+    int32_t fit_params_min_ctx = 4096; // minimum context size to set when trying to reduce memory use
 
     // margin per device in bytes for fitting parameters to free memory:
     std::vector<size_t> fit_params_target = std::vector<size_t>(llama_max_devices(), 1024 * 1024*1024);
 
     enum llama_split_mode split_mode = LLAMA_SPLIT_MODE_LAYER; // how to split the model across GPUs
 
-    common_cpu_params cpuparams;
-    common_cpu_params cpuparams_batch;
+    struct cpu_params cpuparams;
+    struct cpu_params cpuparams_batch;
 
     ggml_backend_sched_eval_callback cb_eval = nullptr;
     void * cb_eval_user_data                 = nullptr;
@@ -479,7 +453,7 @@ struct common_params {
 
     std::set<std::string> model_alias;     // model aliases                                                 // NOLINT
     std::set<std::string> model_tags;      // model tags (informational, not used for routing)              // NOLINT
-    std::string hf_token             = ""; // HF token (aka bearer token)                                   // NOLINT
+    std::string hf_token             = ""; // HF token                                                      // NOLINT
     std::string prompt               = "";                                                                  // NOLINT
     std::string system_prompt        = "";                                                                  // NOLINT
     std::string prompt_file          = ""; // store the external prompt file name                           // NOLINT
@@ -507,7 +481,6 @@ struct common_params {
     int32_t control_vector_layer_start = -1; // layer range for control vector
     int32_t control_vector_layer_end   = -1; // layer range for control vector
     bool    offline                    = false;
-    bool    skip_download              = false; // skip model file downloading
 
     int32_t ppl_stride      = 0;     // stride for perplexity calculations. If left at 0, the pre-existing approach will be used.
     int32_t ppl_output_type = 0;     // = 0 -> ppl output is as usual, = 1 -> ppl output is num_tokens, ppl, one per line
@@ -588,14 +561,14 @@ struct common_params {
     // server params
     int32_t port                = 8080;          // server listens on this network port
     bool    reuse_port          = false;         // allow multiple sockets to bind to the same port
-    int32_t timeout_read        = 3600;          // http read timeout in seconds
+    int32_t timeout_read        = 600;           // http read timeout in seconds
     int32_t timeout_write       = timeout_read;  // http write timeout in seconds
     int32_t n_threads_http      = -1;    // number of threads to process HTTP requests (TODO: support threadpool)
     int32_t n_cache_reuse       = 0;     // min chunk size to reuse from the cache via KV shifting
     bool    cache_prompt        = true;  // whether to enable prompt caching
     bool    cache_idle_slots    = true;  // save and clear idle slots upon starting a new task
     int32_t n_ctx_checkpoints   = 32;    // max number of context checkpoints per slot
-    int32_t checkpoint_min_step = 256;   // minimum spacing between context checkpoints
+    int32_t checkpoint_every_nt = 8192;  // make a checkpoint every n tokens during prefill
     int32_t cache_ram_mib       = 8192;  // -1 = no limit, 0 - disable, 1 = 1 MiB, etc.
 
     std::string hostname      = "127.0.0.1";
@@ -607,6 +580,8 @@ struct common_params {
     bool force_pure_content_parser = false;
     common_reasoning_format reasoning_format = COMMON_REASONING_FORMAT_DEEPSEEK;
     int enable_reasoning = -1; // -1 = auto, 0 = disable, 1 = enable
+    int reasoning_budget = -1;
+    std::string reasoning_budget_message; // message injected before end tag when budget exhausted
     bool prefill_assistant = true; // if true, any trailing assistant message will be prefilled into the response
     int sleep_idle_seconds = -1;   // if >0, server will sleep after this many seconds of idle time
 
@@ -617,17 +592,11 @@ struct common_params {
 
     std::map<std::string, std::string> default_template_kwargs;
 
-    // UI configs
-    bool ui = true;
-
-    // Deprecated: use ui, ui_mcp_proxy, ui_config_json instead
-    bool webui = ui;
+    // webui configs
+    bool webui = true;
     bool webui_mcp_proxy = false;
     std::string webui_config_json;
 
-    bool ui_mcp_proxy = false;
-    std::string ui_config_json;
-
     // "advanced" endpoints are disabled by default for better security
     bool endpoint_slots   = true;
     bool endpoint_props   = false; // only control POST requests, not GET
@@ -705,12 +674,11 @@ struct common_params {
 // initializes the logging system and prints info about the build
 void common_init();
 
-void common_params_print_info(const common_params & params, bool print_devices = true);
 std::string common_params_get_system_info(const common_params & params);
 
 bool parse_cpu_range(const std::string & range, bool(&boolmask)[GGML_MAX_N_THREADS]);
 bool parse_cpu_mask(const std::string & mask, bool(&boolmask)[GGML_MAX_N_THREADS]);
-void postprocess_cpu_params(common_cpu_params & cpuparams, const common_cpu_params * role_model = nullptr);
+void postprocess_cpu_params(cpu_params & cpuparams, const cpu_params * role_model = nullptr);
 bool set_process_priority(enum ggml_sched_priority prio);
 
 //
@@ -732,7 +700,6 @@ std::string string_format(const char * fmt, ...);
 
 std::string string_strip(const std::string & str);
 std::string string_get_sortable_timestamp();
-std::string string_lcs(std::string_view a, std::string_view b);
 
 std::string string_join(const std::vector<std::string> & values, const std::string & separator);
 std::vector<std::string> string_split(const std::string & str, const std::string & delimiter);
@@ -779,11 +746,6 @@ inline bool string_starts_with(std::string_view str, std::string_view prefix) {
            str.compare(0, prefix.size(), prefix) == 0;
 }
 
-// remove when moving to c++20
-inline bool string_starts_with(std::string_view str, char prefix) {
-    return !str.empty() && str.front() == prefix;
-}
-
 // remove when moving to c++20
 inline bool string_ends_with(std::string_view str, std::string_view suffix) {
     return str.size() >= suffix.size() &&
@@ -857,7 +819,7 @@ struct common_sampler;
 
 // note: defines the model, context, samplers, ets. lifetimes
 struct common_init_result {
-    common_init_result(common_params & params, bool model_only = false);
+    common_init_result(common_params & params);
     ~common_init_result();
 
     llama_model * model();
@@ -875,11 +837,11 @@ private:
 
 using common_init_result_ptr = std::unique_ptr<common_init_result>;
 
-common_init_result_ptr common_init_from_params(common_params & params, bool model_only = false);
+common_init_result_ptr common_init_from_params(common_params & params);
 
 struct llama_model_params     common_model_params_to_llama  (      common_params & params);
 struct llama_context_params   common_context_params_to_llama(const common_params & params);
-struct ggml_threadpool_params ggml_threadpool_params_from_cpu_params(const common_cpu_params & params);
+struct ggml_threadpool_params ggml_threadpool_params_from_cpu_params(const cpu_params & params);
 
 // clear LoRA adapters from context, then apply new list of adapters
 void common_set_adapter_lora(struct llama_context * ctx, std::vector<common_adapter_lora_info> & lora);
@@ -892,20 +854,15 @@ std::string common_get_model_endpoint();
 //
 
 enum common_context_seq_rm_type {
-    COMMON_CONTEXT_SEQ_RM_TYPE_NO           = 0, // seq_rm not supported (e.g. no memory module)
-    COMMON_CONTEXT_SEQ_RM_TYPE_PART         = 1, // can seq_rm partial sequences
-    COMMON_CONTEXT_SEQ_RM_TYPE_FULL         = 2, // can seq_rm full sequences only
-    COMMON_CONTEXT_SEQ_RM_TYPE_RS = 3, // can seq_rm partial sequences, bounded by n_rs_seq
+    COMMON_CONTEXT_SEQ_RM_TYPE_NO   = 0, // seq_rm not supported (e.g. no memory module)
+    COMMON_CONTEXT_SEQ_RM_TYPE_PART = 1, // can seq_rm partial sequences
+    COMMON_CONTEXT_SEQ_RM_TYPE_FULL = 2, // can seq_rm full sequences only
 };
 
 // check if the llama_context can remove sequences
 // note: clears the memory of the context
 common_context_seq_rm_type common_context_can_seq_rm(llama_context * ctx);
 
-// aborts execution on failure
-void common_context_seq_rm (llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1);
-void common_context_seq_add(llama_context * ctx, llama_seq_id seq_id, llama_pos p0, llama_pos p1, llama_pos delta);
-void common_context_seq_cp (llama_context * ctx, llama_seq_id seq_id_src, llama_seq_id seq_id_dst, llama_pos p0, llama_pos p1);
 
 //
 // Batch utils
@@ -1044,50 +1001,3 @@ ggml_opt_dataset_t common_opt_dataset_init(struct llama_context * ctx, const std
 
 // "adamw" or "sgd" (case insensitive)
 enum ggml_opt_optimizer_type common_opt_get_optimizer(const char *);
-
-//
-// prompt utils
-//
-
-struct common_prompt_checkpoint {
-    int64_t n_tokens;
-
-    llama_pos pos_min;
-    llama_pos pos_max;
-
-    std::vector<uint8_t> data_tgt;
-    std::vector<uint8_t> data_dft;
-
-    size_t size() const;
-
-    bool empty() const;
-    void clear();
-
-    void update_pos(
-            int64_t n_tokens,
-            llama_pos pos_min,
-            llama_pos pos_max);
-
-    void update_tgt(
-            llama_context * ctx,
-            llama_seq_id seq_id,
-            llama_state_seq_flags flags);
-
-    void update_dft(
-            llama_context * ctx,
-            llama_seq_id seq_id,
-            llama_state_seq_flags flags);
-
-    void load_tgt(
-            llama_context * ctx,
-            llama_seq_id seq_id,
-            llama_state_seq_flags flags) const;
-
-    void load_dft(
-            llama_context * ctx,
-            llama_seq_id seq_id,
-            llama_state_seq_flags flags) const;
-
-    void clear_tgt();
-    void clear_dft();
-};

common/debug.cpp

@@ -1,38 +1,9 @@
 #include "debug.h"
 
-#include "common.h"
 #include "log.h"
 
 #include <cmath>
-#include <regex>
 #include <string>
-#include <vector>
-
-struct common_debug_cb_user_data::impl {
-    std::vector<uint8_t>    data;
-    std::vector<std::regex> tensor_filters;
-    bool                    abort_on_nan{false};
-};
-
-common_debug_cb_user_data::common_debug_cb_user_data() : pimpl(std::make_unique<impl>()) {}
-common_debug_cb_user_data::~common_debug_cb_user_data() = default;
-
-common_debug_cb_user_data::common_debug_cb_user_data(common_params & params, const std::vector<std::string> & filter_patterns, bool abort_on_nan)
-    : pimpl(std::make_unique<impl>())
-{
-    for (const auto & pattern : filter_patterns) {
-        try {
-            std::string anchored_pattern = "^" + pattern;
-            pimpl->tensor_filters.emplace_back(anchored_pattern, std::regex::optimize);
-        } catch (const std::regex_error & e) {
-            throw std::runtime_error("Invalid regex pattern '" + pattern + "': " + e.what());
-        }
-    }
-    pimpl->abort_on_nan = abort_on_nan;
-
-    params.cb_eval           = common_debug_cb_eval;
-    params.cb_eval_user_data = this;
-}
 
 static std::string common_ggml_ne_string(const ggml_tensor * t) {
     std::string str;
@@ -76,7 +47,8 @@ static float common_ggml_get_float_value(const uint8_t * data,
 
 #define INDENT "    "
 
-static void common_debug_print_tensor(uint8_t * data, ggml_type type, const int64_t * ne, const size_t * nb, int64_t n, bool abort_on_nan) {
+template <bool abort>
+void common_debug_print_tensor(uint8_t * data, ggml_type type, const int64_t * ne, const size_t * nb, int64_t n) {
     GGML_ASSERT(n > 0);
     float sum = 0;
     for (int64_t i3 = 0; i3 < ne[3]; i3++) {
@@ -122,7 +94,7 @@ static void common_debug_print_tensor(uint8_t * data, ggml_type type, const int6
         LOG(INDENT "sum = %f\n", sum);
     }
 
-    if (abort_on_nan) {
+    if constexpr (abort) {
         if (std::isnan(sum)) {
             LOG("encountered NaN - aborting\n");
             exit(0);
@@ -140,9 +112,8 @@ static void common_debug_print_tensor(uint8_t * data, ggml_type type, const int6
  * @param user_data user data to pass at each call back
  * @return true to receive data or continue the graph, false otherwise
  */
-bool common_debug_cb_eval(struct ggml_tensor * t, bool ask, void * user_data) {
-    auto * cb_data = (common_debug_cb_user_data *) user_data;
-    auto * pimpl = cb_data->pimpl.get();
+template <bool abort_on_nan> bool common_debug_cb_eval(struct ggml_tensor * t, bool ask, void * user_data) {
+    auto * cb_data = (base_callback_data *) user_data;
 
     const struct ggml_tensor * src0 = t->src[0];
     const struct ggml_tensor * src1 = t->src[1];
@@ -151,10 +122,10 @@ bool common_debug_cb_eval(struct ggml_tensor * t, bool ask, void * user_data) {
         return true;  // Always retrieve data
     }
 
-    bool matches_filter = pimpl->tensor_filters.empty();
+    bool matches_filter = cb_data->tensor_filters.empty();
 
     if (!matches_filter) {
-        for (const auto & filter : pimpl->tensor_filters) {
+        for (const auto & filter : cb_data->tensor_filters) {
             if (std::regex_search(t->name, filter)) {
                 matches_filter = true;
                 break;
@@ -177,14 +148,20 @@ bool common_debug_cb_eval(struct ggml_tensor * t, bool ask, void * user_data) {
 
     if (!is_host) {
         auto n_bytes = ggml_nbytes(t);
-        pimpl->data.resize(n_bytes);
-        ggml_backend_tensor_get(t, pimpl->data.data(), 0, n_bytes);
+        cb_data->data.resize(n_bytes);
+        ggml_backend_tensor_get(t, cb_data->data.data(), 0, n_bytes);
     }
 
     if (!ggml_is_quantized(t->type) && matches_filter) {
-        uint8_t * data = is_host ? (uint8_t *) t->data : pimpl->data.data();
-        common_debug_print_tensor(data, t->type, t->ne, t->nb, 3, pimpl->abort_on_nan);
+        uint8_t * data = is_host ? (uint8_t *) t->data : cb_data->data.data();
+        common_debug_print_tensor<abort_on_nan>(data, t->type, t->ne, t->nb, 3);
     }
 
     return true;
 }
+
+// Explicit template instantiations
+template bool common_debug_cb_eval<false>(ggml_tensor *, bool, void *);
+template bool common_debug_cb_eval<true>(ggml_tensor *, bool, void *);
+template void common_debug_print_tensor<false>(uint8_t *, ggml_type, const int64_t *, const size_t *, int64_t);
+template void common_debug_print_tensor<true>(uint8_t *, ggml_type, const int64_t *, const size_t *, int64_t);

common/debug.h

@@ -1,31 +1,43 @@
 #pragma once
-
-#include <memory>
+#include "common.h"
 #include <string>
 #include <vector>
+#include <regex>
 
 // common debug functions and structs
 
-struct common_params;
+// Print a tensor's detailed data
+// data - the tensor's data in byte format
+// type - the tensor's quantization type
+// ne   - the tensor dimensions array
+// nb   - the tensor strides array
+// n    - the number of rows/columns to fully print
+template <bool abort_on_nan> void common_debug_print_tensor(uint8_t * data, ggml_type type, const int64_t * ne, const size_t * nb, int64_t n);
 
 // Intended to use as callback for ggml_backend_sched_eval_callback
 // prints tensors that are processed in the computation graph
-// by default prints all tensors, but can be configured by creating a `common_debug_cb_user_data` instance with
-// non-empty filter_patterns. See examples/debug.cpp for possible usage patterns
-// `common_debug_cb_user_data` contains `abort_on_nan` flag that determines whether an error should be thrown whenever a NaN is encountered
+// by default prints all tensors, but can be configured by creating a `base_callback_data` instance with
+// non-empty filter_patterns. See examples/debug.ccp for possible usage patterns
+// The template parameter determines whether an error should be thrown whenever a NaN is encountered
 // in a tensor (useful for stopping debug sessions on first erroneous tensor)
 // The callback data will be passed as the third parameter (user_data)
-bool common_debug_cb_eval(struct ggml_tensor * t, bool ask, void * user_data);
+template <bool abort_on_nan> bool common_debug_cb_eval(struct ggml_tensor * t, bool ask, void * user_data);
+struct base_callback_data {
+    std::vector<uint8_t>    data;
+    std::vector<std::regex> tensor_filters;
 
-struct common_debug_cb_user_data {
-    struct impl;
-    std::unique_ptr<impl> pimpl;
+    base_callback_data() = default;
 
-    common_debug_cb_user_data();
-    ~common_debug_cb_user_data();
-
-    common_debug_cb_user_data(const common_debug_cb_user_data &) = delete;
-    common_debug_cb_user_data & operator=(const common_debug_cb_user_data &) = delete;
-
-    common_debug_cb_user_data(common_params & params, const std::vector<std::string> & filter_patterns, bool abort_on_nan = false);
+    base_callback_data(common_params & params, const std::vector<std::string> & filter_patterns) {
+        for (const auto & pattern : filter_patterns) {
+            try {
+                std::string anchored_pattern = "^" + pattern;
+                tensor_filters.emplace_back(anchored_pattern, std::regex::optimize);
+            } catch (const std::regex_error & e) {
+                throw std::runtime_error("Invalid regex pattern '" + pattern + "': " + e.what());
+            }
+        }
+        params.cb_eval           = common_debug_cb_eval<false>;
+        params.cb_eval_user_data = this;
+    }
 };

common/download.cpp

@@ -292,10 +292,6 @@ static int common_download_file_single_online(const std::string & url,
 
     const bool file_exists = std::filesystem::exists(path);
 
-    if (!file_exists && opts.skip_download) {
-        return -2; // file is missing and download is disabled
-    }
-
     if (file_exists && skip_etag) {
         LOG_DBG("%s: using cached file: %s\n", __func__, path.c_str());
         return 304; // 304 Not Modified - fake cached response
@@ -324,9 +320,9 @@ static int common_download_file_single_online(const std::string & url,
 
     auto head = cli.Head(parts.path);
     if (!head || head->status < 200 || head->status >= 300) {
-        LOG_TRC("%s: HEAD failed, status: %d\n", __func__, head ? head->status : -1);
+        LOG_WRN("%s: HEAD failed, status: %d\n", __func__, head ? head->status : -1);
         if (file_exists) {
-            LOG_TRC("%s: using cached file (HEAD failed): %s\n", __func__, path.c_str());
+            LOG_INF("%s: using cached file (HEAD failed): %s\n", __func__, path.c_str());
             return 304; // 304 Not Modified - fake cached response
         }
         return head ? head->status : -1;
@@ -361,10 +357,6 @@ static int common_download_file_single_online(const std::string & url,
             LOG_DBG("%s: using cached file (same etag): %s\n", __func__, path.c_str());
             return 304; // 304 Not Modified - fake cached response
         }
-        // pass this point, the file exists but is different from the server version, so we need to redownload it
-        if (opts.skip_download) {
-            return -2; // special code to indicate that the download was skipped due to etag mismatch
-        }
         if (remove(path.c_str()) != 0) {
             LOG_ERR("%s: unable to delete file: %s\n", __func__, path.c_str());
             return -1;
@@ -574,11 +566,8 @@ static hf_cache::hf_files get_split_files(const hf_cache::hf_files & files,
     return result;
 }
 
-// pick the best sibling GGUF whose filename contains `keyword` (e.g. "mmproj" / "mtp"),
-// preferring deeper shared directory prefix with the model, then closest quantization
-static hf_cache::hf_file find_best_sibling(const hf_cache::hf_files & files,
-                                           const std::string        & model,
-                                           const std::string        & keyword) {
+static hf_cache::hf_file find_best_mmproj(const hf_cache::hf_files & files,
+                                          const std::string        & model) {
     hf_cache::hf_file best;
     size_t best_depth = 0;
     int best_diff = 0;
@@ -590,20 +579,20 @@ static hf_cache::hf_file find_best_sibling(const hf_cache::hf_files & files,
 
     for (const auto & f : files) {
         if (!string_ends_with(f.path, ".gguf") ||
-            f.path.find(keyword) == std::string::npos) {
+            f.path.find("mmproj") == std::string::npos) {
             continue;
         }
 
-        auto sib_parts = string_split<std::string>(f.path, '/');
-        auto sib_dir = sib_parts.end() - 1;
+        auto mmproj_parts = string_split<std::string>(f.path, '/');
+        auto mmproj_dir = mmproj_parts.end() - 1;
 
         auto [_, dir] = std::mismatch(model_parts.begin(), model_dir,
-                                      sib_parts.begin(), sib_dir);
-        if (dir != sib_dir) {
+                                      mmproj_parts.begin(), mmproj_dir);
+        if (dir != mmproj_dir) {
             continue;
         }
 
-        size_t depth = dir - sib_parts.begin();
+        size_t depth = dir - mmproj_parts.begin();
         auto bits = extract_quant_bits(f.path);
         auto diff = std::abs(bits - model_bits);
 
@@ -617,16 +606,6 @@ static hf_cache::hf_file find_best_sibling(const hf_cache::hf_files & files,
     return best;
 }
 
-static hf_cache::hf_file find_best_mmproj(const hf_cache::hf_files & files,
-                                          const std::string        & model) {
-    return find_best_sibling(files, model, "mmproj");
-}
-
-static hf_cache::hf_file find_best_mtp(const hf_cache::hf_files & files,
-                                       const std::string        & model) {
-    return find_best_sibling(files, model, "mtp-");
-}
-
 static bool gguf_filename_is_model(const std::string & filepath) {
     if (!string_ends_with(filepath, ".gguf")) {
         return false;
@@ -638,8 +617,7 @@ static bool gguf_filename_is_model(const std::string & filepath) {
     }
 
     return filename.find("mmproj")  == std::string::npos &&
-           filename.find("imatrix") == std::string::npos &&
-           filename.find("mtp-")    == std::string::npos;
+           filename.find("imatrix") == std::string::npos;
 }
 
 static hf_cache::hf_file find_best_model(const hf_cache::hf_files & files,
@@ -649,7 +627,7 @@ static hf_cache::hf_file find_best_model(const hf_cache::hf_files & files,
     if (!tag.empty()) {
         tags.push_back(tag);
     } else {
-        tags = {"Q4_K_M", "Q8_0"};
+        tags = {"Q4_K_M", "Q4_0"};
     }
 
     for (const auto & t : tags) {
@@ -695,13 +673,11 @@ struct hf_plan {
     hf_cache::hf_file primary;
     hf_cache::hf_files model_files;
     hf_cache::hf_file mmproj;
-    hf_cache::hf_file mtp;
 };
 
 static hf_plan get_hf_plan(const common_params_model  & model,
                            const common_download_opts & opts,
-                           bool download_mmproj,
-                           bool download_mtp) {
+                           bool download_mmproj) {
     hf_plan plan;
     hf_cache::hf_files all;
 
@@ -747,10 +723,6 @@ static hf_plan get_hf_plan(const common_params_model  & model,
         plan.mmproj = find_best_mmproj(all, primary.path);
     }
 
-    if (download_mtp) {
-        plan.mtp = find_best_mtp(all, primary.path);
-    }
-
     return plan;
 }
 
@@ -783,26 +755,22 @@ static std::vector<download_task> get_url_tasks(const common_params_model & mode
 }
 
 common_download_model_result common_download_model(const common_params_model  & model,
-                                                   const common_download_opts & opts) {
+                                                   const common_download_opts & opts,
+                                                   bool download_mmproj) {
     common_download_model_result result;
     std::vector<download_task> tasks;
     hf_plan hf;
 
-    bool download_mmproj = opts.download_mmproj;
-    bool download_mtp = opts.download_mtp;
     bool is_hf = !model.hf_repo.empty();
 
     if (is_hf) {
-        hf = get_hf_plan(model, opts, download_mmproj, download_mtp);
+        hf = get_hf_plan(model, opts, download_mmproj);
         for (const auto & f : hf.model_files) {
             tasks.push_back({f.url, f.local_path});
         }
         if (!hf.mmproj.path.empty()) {
             tasks.push_back({hf.mmproj.url, hf.mmproj.local_path});
         }
-        if (!hf.mtp.path.empty()) {
-            tasks.push_back({hf.mtp.url, hf.mtp.local_path});
-        }
     } else if (!model.url.empty()) {
         tasks = get_url_tasks(model);
     } else {
@@ -814,22 +782,18 @@ common_download_model_result common_download_model(const common_params_model  &
         return result;
     }
 
-    std::vector<std::future<int>> futures;
+    std::vector<std::future<bool>> futures;
     for (const auto & task : tasks) {
         futures.push_back(std::async(std::launch::async,
             [&task, &opts, is_hf]() {
-                return common_download_file_single(task.url, task.path, opts, is_hf);
+                int status = common_download_file_single(task.url, task.path, opts, is_hf);
+                return is_http_status_ok(status);
             }
         ));
     }
 
     for (auto & f : futures) {
-        int status = f.get();
-        if (status == -2 && opts.skip_download) {
-            throw common_skip_download_exception();
-        }
-        bool is_ok = is_http_status_ok(status);
-        if (!is_ok) {
+        if (!f.get()) {
             return {};
         }
     }
@@ -843,10 +807,6 @@ common_download_model_result common_download_model(const common_params_model  &
         if (!hf.mmproj.path.empty()) {
             result.mmproj_path = hf_cache::finalize_file(hf.mmproj);
         }
-
-        if (!hf.mtp.path.empty()) {
-            result.mtp_path = hf_cache::finalize_file(hf.mtp);
-        }
     } else {
         result.model_path = model.path;
     }
@@ -986,8 +946,7 @@ std::vector<common_cached_model_info> common_list_cached_models() {
     for (const auto & f : files) {
         auto split = get_gguf_split_info(f.path);
         if (split.index != 1 || split.tag.empty() ||
-            split.prefix.find("mmproj") != std::string::npos ||
-            split.prefix.find("mtp-")   != std::string::npos) {
+            split.prefix.find("mmproj") != std::string::npos) {
             continue;
         }
         if (seen.insert(f.repo_id + ":" + split.tag).second) {

common/download.h

@@ -52,9 +52,6 @@ struct common_download_opts {
     std::string bearer_token;
     common_header_list headers;
     bool offline = false;
-    bool skip_download = false; // if true, only validation is performed, common_skip_download_exception may be thrown if the file is missing or invalid
-    bool download_mmproj = false;
-    bool download_mtp = false;
     common_download_callback * callback = nullptr;
 };
 
@@ -62,12 +59,6 @@ struct common_download_opts {
 struct common_download_model_result {
     std::string model_path;
     std::string mmproj_path;
-    std::string mtp_path;
-};
-
-// throw if the file is missing or invalid (e.g. ETag check failed)
-struct common_skip_download_exception : public std::runtime_error {
-    common_skip_download_exception() : std::runtime_error("skip download") {}
 };
 
 // Download model from HuggingFace repo or URL
@@ -92,12 +83,12 @@ struct common_skip_download_exception : public std::runtime_error {
 // when opts.offline=true, no network requests are made
 // when download_mmproj=true, searches for mmproj in same directory as model or any parent directory
 // then with the closest quantization bits
-// when download_mtp=true, applies the same sibling search for an MTP-head GGUF
 //
-// returns result with model_path, mmproj_path and mtp_path (empty when not found / on failure)
+// returns result with model_path and mmproj_path (empty on failure)
 common_download_model_result common_download_model(
     const common_params_model & model,
-    const common_download_opts & opts = {}
+    const common_download_opts & opts = {},
+    bool download_mmproj = false
 );
 
 // returns list of cached models
@@ -105,7 +96,6 @@ std::vector<common_cached_model_info> common_list_cached_models();
 
 // download single file from url to local path
 // returns status code or -1 on error
-// returns -2 if the download was skipped due to ETag mismatch (file outdated, skip_download=true)
 // skip_etag: if true, don't read/write .etag files (for HF cache where filename is the hash)
 int common_download_file_single(const std::string & url,
                                 const std::string & path,

common/fit.cpp

@@ -1,959 +0,0 @@
-#include "fit.h"
-
-#include "log.h"
-
-#include "../src/llama-ext.h"
-
-#include <array>
-#include <cassert>
-#include <stdexcept>
-#include <cinttypes>
-#include <set>
-#include <string>
-#include <vector>
-
-// this enum is only used in llama_params_fit_impl but needs to be defined outside of it to fix a Windows compilation issue
-// enum to identify part of a layer for distributing its tensors:
-enum common_layer_fraction_t {
-    LAYER_FRACTION_NONE = 0, // nothing
-    LAYER_FRACTION_ATTN = 1, // attention
-    LAYER_FRACTION_UP   = 2, // attention + up
-    LAYER_FRACTION_GATE = 3, // attention + up + gate
-    LAYER_FRACTION_MOE  = 4, // everything but sparse MoE weights
-};
-
-class common_params_fit_exception : public std::runtime_error {
-    using std::runtime_error::runtime_error;
-};
-
-std::vector<llama_device_memory_data> common_get_device_memory_data(
-        const char * path_model,
-        const llama_model_params * mparams,
-        const llama_context_params * cparams,
-        std::vector<ggml_backend_dev_t> & devs,
-        uint32_t & hp_ngl,
-        uint32_t & hp_n_ctx_train,
-        uint32_t & hp_n_expert,
-        ggml_log_level log_level) {
-    struct user_data_t {
-        struct {
-            ggml_log_callback callback;
-            void * user_data;
-        } original_logger;
-        ggml_log_level min_level; // prints below this log level go to debug log
-    };
-    user_data_t ud;
-    llama_log_get(&ud.original_logger.callback, &ud.original_logger.user_data);
-    ud.min_level = log_level;
-
-    llama_log_set([](ggml_log_level level, const char * text, void * user_data) {
-        const user_data_t * ud = (const user_data_t *) user_data;
-        const ggml_log_level level_eff = level >= ud->min_level ? level : GGML_LOG_LEVEL_DEBUG;
-        ud->original_logger.callback(level_eff, text, ud->original_logger.user_data);
-    }, &ud);
-
-    llama_model_params mparams_copy = *mparams;
-    mparams_copy.no_alloc  = true;
-    mparams_copy.use_mmap  = false;
-    mparams_copy.use_mlock = false;
-
-    llama_model * model = llama_model_load_from_file(path_model, mparams_copy);
-    if (model == nullptr) {
-        llama_log_set(ud.original_logger.callback, ud.original_logger.user_data);
-        throw std::runtime_error("failed to load model");
-    }
-
-    llama_context * ctx = llama_init_from_model(model, *cparams);
-    if (ctx == nullptr) {
-        llama_model_free(model);
-        llama_log_set(ud.original_logger.callback, ud.original_logger.user_data);
-        throw std::runtime_error("failed to create llama_context from model");
-    }
-
-    const size_t nd = llama_model_n_devices(model);
-    std::vector<llama_device_memory_data> ret(nd + 1);
-
-    llama_memory_breakdown memory_breakdown = llama_get_memory_breakdown(ctx);
-
-    for (const auto & [buft, mb] : memory_breakdown) {
-        if (ggml_backend_buft_is_host(buft)) {
-            ret.back().mb.model   += mb.model;
-            ret.back().mb.context += mb.context;
-            ret.back().mb.compute += mb.compute;
-            continue;
-        }
-
-        ggml_backend_dev_t dev = ggml_backend_buft_get_device(buft);
-        if (!dev) {
-            continue;
-        }
-        for (size_t i = 0; i < nd; i++) {
-            if (dev == llama_model_get_device(model, i)) {
-                ret[i].mb.model   += mb.model;
-                ret[i].mb.context += mb.context;
-                ret[i].mb.compute += mb.compute;
-                break;
-            }
-        }
-    }
-
-    {
-        ggml_backend_dev_t cpu_dev = ggml_backend_dev_by_type(GGML_BACKEND_DEVICE_TYPE_CPU);
-        if (cpu_dev == nullptr) {
-            throw std::runtime_error("no CPU backend found");
-        }
-        size_t free;
-        size_t total;
-        ggml_backend_dev_memory(cpu_dev, &free, &total);
-        ret.back().free  = free;
-        ret.back().total = total;
-    }
-    for (size_t i = 0; i < nd; i++) {
-        ggml_backend_dev_t dev = llama_model_get_device(model, i);
-
-        size_t free;
-        size_t total;
-        ggml_backend_dev_memory(dev, &free, &total);
-
-        // Some non-GPU accelerator backends, such as BLAS, report 0/0 and rely on
-        // the host-memory fallback. For GPU-like backends, keep 0/0 so --fit does
-        // not assign anything to a device with an unknown memory budget.
-        if (free == 0 && total == 0) {
-            const enum ggml_backend_dev_type type = ggml_backend_dev_type(dev);
-            if (type == GGML_BACKEND_DEVICE_TYPE_GPU || type == GGML_BACKEND_DEVICE_TYPE_IGPU) {
-                LOG_WRN("%s: device %s did not report memory; --fit will not use it\n",
-                        __func__, ggml_backend_dev_name(dev));
-            } else {
-                free  = ret.back().free;
-                total = ret.back().total;
-            }
-        }
-        ret[i].free  = free;
-        ret[i].total = total;
-    }
-
-    devs.clear();
-    for (int i = 0; i < llama_model_n_devices(model); i++) {
-        devs.push_back(llama_model_get_device(model, i));
-    }
-
-    hp_ngl         = llama_model_n_layer(model);
-    hp_n_ctx_train = llama_model_n_ctx_train(model);
-    hp_n_expert    = llama_model_n_expert(model);
-
-    common_memory_breakdown_print(ctx);
-
-    llama_free(ctx);
-    llama_model_free(model);
-    llama_log_set(ud.original_logger.callback, ud.original_logger.user_data);
-
-    return ret;
-}
-
-static void common_params_fit_impl(
-        const char * path_model, struct llama_model_params * mparams, struct llama_context_params * cparams,
-        float * tensor_split, struct llama_model_tensor_buft_override * tensor_buft_overrides,
-        size_t * margins_s, uint32_t n_ctx_min, enum ggml_log_level log_level) {
-    if (mparams->split_mode == LLAMA_SPLIT_MODE_TENSOR) {
-        throw common_params_fit_exception("llama_params_fit is not implemented for SPLIT_MODE_TENSOR, abort");
-    }
-    constexpr int64_t MiB = 1024*1024;
-    typedef std::vector<llama_device_memory_data> dmds_t;
-    const llama_model_params default_mparams = llama_model_default_params();
-
-    std::vector<ggml_backend_dev_t> devs;
-    uint32_t hp_ngl = 0; // hparams.n_gpu_layers
-    uint32_t hp_nct = 0; // hparams.n_ctx_train
-    uint32_t hp_nex = 0; // hparams.n_expert
-
-    // step 1: get data for default parameters and check whether any changes are necessary in the first place
-
-    LOG_TRC("%s: getting device memory data for initial parameters:\n", __func__);
-    const dmds_t dmds_full = common_get_device_memory_data(path_model, mparams, cparams, devs, hp_ngl, hp_nct, hp_nex, log_level);
-    const size_t nd = devs.size(); // number of devices
-
-    std::vector<int64_t> margins; // this function uses int64_t rather than size_t for memory sizes to more conveniently handle deficits
-    margins.reserve(nd);
-    if (nd == 0) {
-        margins.push_back(margins_s[0]);
-    } else {
-        for (size_t id = 0; id < nd; id++) {
-            margins.push_back(margins_s[id]);
-        }
-    }
-
-    std::vector<std::string> dev_names;
-    {
-        dev_names.reserve(nd);
-        size_t max_length = 0;
-        for (const auto & dev : devs) {
-            std::string name = ggml_backend_dev_name(dev);
-            name += " (";
-            name += ggml_backend_dev_description(dev);
-            name += ")";
-            dev_names.push_back(name);
-            max_length = std::max(max_length, name.length());
-        }
-        for (std::string & dn : dev_names) {
-            dn.insert(dn.end(), max_length - dn.length(), ' ');
-        }
-    }
-
-    int64_t sum_free            = 0;
-    int64_t sum_projected_free  = 0;
-    int64_t sum_projected_used  = 0;
-    int64_t sum_projected_model = 0;
-    std::vector<int64_t> projected_free_per_device;
-    projected_free_per_device.reserve(nd);
-
-    if (nd == 0) {
-        sum_projected_used = dmds_full.back().mb.total();
-        sum_free           = dmds_full.back().total;
-        sum_projected_free = sum_free - sum_projected_used;
-        LOG_INF("%s: projected to use %" PRId64 " MiB of host memory vs. %" PRId64 " MiB of total host memory\n",
-            __func__, sum_projected_used/MiB, sum_free/MiB);
-        if (sum_projected_free >= margins[0]) {
-            LOG_TRC("%s: will leave %" PRId64 " >= %" PRId64 " MiB of system memory, no changes needed\n",
-                __func__, sum_projected_free/MiB, margins[0]/MiB);
-            return;
-        }
-    } else {
-        if (nd > 1) {
-            LOG_TRC("%s: projected memory use with initial parameters [MiB]:\n", __func__);
-        }
-        for (size_t id = 0; id < nd; id++) {
-            const llama_device_memory_data & dmd = dmds_full[id];
-
-            const int64_t projected_used = dmd.mb.total();
-            const int64_t projected_free = dmd.free - projected_used;
-            projected_free_per_device.push_back(projected_free);
-
-            sum_free            += dmd.free;
-            sum_projected_used  += projected_used;
-            sum_projected_free  += projected_free;
-            sum_projected_model += dmd.mb.model;
-
-            if (nd > 1) {
-                LOG_TRC("%s:   - %s: %6" PRId64 " total, %6" PRId64 " used, %6" PRId64 " free vs. target of %6" PRId64 "\n",
-                    __func__, dev_names[id].c_str(), dmd.total/MiB, projected_used/MiB, projected_free/MiB, margins[id]/MiB);
-            }
-        }
-        assert(sum_free >= 0 && sum_projected_used >= 0);
-        LOG_TRC("%s: projected to use %" PRId64 " MiB of device memory vs. %" PRId64 " MiB of free device memory\n",
-            __func__, sum_projected_used/MiB, sum_free/MiB);
-        if (nd == 1) {
-            if (projected_free_per_device[0] >= margins[0]) {
-                LOG_TRC("%s: will leave %" PRId64 " >= %" PRId64 " MiB of free device memory, no changes needed\n",
-                    __func__, projected_free_per_device[0]/MiB, margins[0]/MiB);
-                return;
-            }
-        } else {
-            bool changes_needed = false;
-            for (size_t id = 0; id < nd; id++) {
-                if (projected_free_per_device[id] < margins[id]) {
-                    changes_needed = true;
-                    break;
-                }
-            }
-            if (!changes_needed) {
-                LOG_TRC("%s: targets for free memory can be met on all devices, no changes needed\n", __func__);
-                return;
-            }
-        }
-    }
-
-    // step 2: try reducing memory use by reducing the context size
-
-    {
-        int64_t global_surplus = sum_projected_free;
-        if (nd == 0) {
-            global_surplus -= margins[0];
-        } else {
-            for (size_t id = 0; id < nd; id++) {
-                global_surplus -= margins[id];
-            }
-        }
-        if (global_surplus < 0) {
-            if (nd <= 1) {
-                LOG_TRC("%s: cannot meet free memory target of %" PRId64 " MiB, need to reduce device memory by %" PRId64 " MiB\n",
-                    __func__, margins[0]/MiB, -global_surplus/MiB);
-            } else {
-                LOG_TRC(
-                    "%s: cannot meet free memory targets on all devices, need to use %" PRId64 " MiB less in total\n",
-                    __func__, -global_surplus/MiB);
-            }
-            if (cparams->n_ctx == 0) {
-                if (hp_nct > n_ctx_min) {
-                    int64_t sum_used_target = sum_free;
-                    if (nd == 0) {
-                        sum_used_target -= margins[0];
-                    } else {
-                        for (size_t id = 0; id < nd; id++) {
-                            sum_used_target -= margins[id];
-                        }
-                    }
-                    if (nd > 1) {
-                        // for multiple devices we need to be more conservative in terms of how much context we think can fit:
-                        //   - for dense models only whole layers can be assigned to devices
-                        //   - for MoE models only whole tensors can be assigned to devices, which we estimate to be <= 1/3 of a layer
-                        //   - on average we expect a waste of 0.5 layers/tensors per device
-                        //   - use slightly more than the expected average for nd devices to be safe
-                        const int64_t model_per_layer = sum_projected_model / std::min(uint32_t(mparams->n_gpu_layers), hp_ngl);
-                        sum_used_target -= (nd + 1) * model_per_layer / (hp_nex == 0 ? 2 : 6);
-                    }
-
-                    int64_t sum_projected_used_min_ctx = 0;
-                    cparams->n_ctx = n_ctx_min;
-                    const dmds_t dmds_min_ctx = common_get_device_memory_data(path_model, mparams, cparams, devs, hp_ngl, hp_nct, hp_nex, log_level);
-                    if (nd == 0) {
-                        sum_projected_used_min_ctx = dmds_min_ctx.back().mb.total();
-                    } else {
-                        for (size_t id = 0; id < nd; id++) {
-                            sum_projected_used_min_ctx += dmds_min_ctx[id].mb.total();
-                        }
-                    }
-                    if (sum_used_target > sum_projected_used_min_ctx) {
-                        // linear interpolation between minimum and maximum context size:
-                        cparams->n_ctx += (hp_nct - n_ctx_min) * (sum_used_target - sum_projected_used_min_ctx)
-                            / (sum_projected_used - sum_projected_used_min_ctx);
-                        cparams->n_ctx = std::max(cparams->n_ctx - cparams->n_ctx % 256, n_ctx_min); // round down context for CUDA backend
-
-                        const int64_t bytes_per_ctx = (sum_projected_used - sum_projected_used_min_ctx) / (hp_nct - n_ctx_min);
-                        const int64_t memory_reduction = (hp_nct - cparams->n_ctx) * bytes_per_ctx;
-                        LOG_TRC("%s: context size reduced from %" PRIu32 " to %" PRIu32 " -> need %" PRId64 " MiB less memory in total\n",
-                            __func__, hp_nct, cparams->n_ctx, memory_reduction/MiB);
-                        if (nd <= 1) {
-                            LOG_TRC("%s: entire model can be fit by reducing context\n", __func__);
-                            return;
-                        }
-                        LOG_TRC("%s: entire model should be fit across devices by reducing context\n", __func__);
-                    } else {
-                        const int64_t memory_reduction = sum_projected_used - sum_projected_used_min_ctx;
-                        LOG_TRC("%s: context size reduced from %" PRIu32 " to %" PRIu32 " -> need %" PRId64 " MiB less memory in total\n",
-                            __func__, hp_nct, cparams->n_ctx, memory_reduction/MiB);
-                    }
-                } else {
-                    if (n_ctx_min == UINT32_MAX) {
-                        LOG_TRC("%s: user has requested full context size of %" PRIu32 " -> no change\n", __func__, hp_nct);
-                    } else {
-                        LOG_TRC("%s: default model context size is %" PRIu32 " which is <= the min. context size of %" PRIu32 " -> no change\n",
-                            __func__, hp_nct, n_ctx_min);
-                    }
-                }
-            } else {
-                LOG_TRC("%s: context size set by user to %" PRIu32 " -> no change\n", __func__, cparams->n_ctx);
-            }
-        }
-    }
-    if (nd == 0) {
-        throw common_params_fit_exception("was unable to fit model into system memory by reducing context, abort");
-    }
-
-    if (mparams->n_gpu_layers != default_mparams.n_gpu_layers) {
-        throw common_params_fit_exception("n_gpu_layers already set by user to " + std::to_string(mparams->n_gpu_layers) + ", abort");
-    }
-    if (nd > 1) {
-        if (!tensor_split) {
-            throw common_params_fit_exception("did not provide a buffer to write the tensor_split to, abort");
-        }
-        if (mparams->tensor_split) {
-            for (size_t id = 0; id < nd; id++) {
-                if (mparams->tensor_split[id] != 0.0f) {
-                    throw common_params_fit_exception("model_params::tensor_split already set by user, abort");
-                }
-            }
-        }
-        if (mparams->split_mode == LLAMA_SPLIT_MODE_ROW) {
-            throw common_params_fit_exception("changing weight allocation for LLAMA_SPLIT_MODE_ROW not implemented, abort");
-        }
-    }
-    if (!tensor_buft_overrides) {
-        throw common_params_fit_exception("did not provide buffer to set tensor_buft_overrides, abort");
-    }
-    if (mparams->tensor_buft_overrides && (mparams->tensor_buft_overrides->pattern || mparams->tensor_buft_overrides->buft)) {
-        throw common_params_fit_exception("model_params::tensor_buft_overrides already set by user, abort");
-    }
-
-    // step 3: iteratively fill the back to front with "dense" layers
-    //   - for a dense model simply fill full layers, giving each device a contiguous slice of the model
-    //   - for a MoE model, same as dense model but with all MoE tensors in system memory
-
-    // utility function that returns a static C string matching the tensors for a specific layer index and layer fraction:
-    auto get_overflow_pattern = [&](const size_t il, const common_layer_fraction_t lf) -> const char * {
-        constexpr size_t n_strings = 1000;
-        if (il >= n_strings) {
-            throw std::runtime_error("at most " + std::to_string(n_strings) + " model layers are supported");
-        }
-        switch (lf) {
-            case LAYER_FRACTION_ATTN: {
-                static std::array<std::string, n_strings> patterns;
-                if (patterns[il].empty()) {
-                    patterns[il] = "blk\\." + std::to_string(il) + "\\.ffn_(gate|up|gate_up|down).*";
-                }
-                return patterns[il].c_str();
-            }
-            case LAYER_FRACTION_UP: {
-                static std::array<std::string, n_strings> patterns;
-                if (patterns[il].empty()) {
-                    patterns[il] = "blk\\." + std::to_string(il) + "\\.ffn_(gate|gate_up|down).*";
-                }
-                return patterns[il].c_str();
-            }
-            case LAYER_FRACTION_GATE: {
-                static std::array<std::string, n_strings> patterns;
-                if (patterns[il].empty()) {
-                    patterns[il] = "blk\\." + std::to_string(il) + "\\.ffn_down.*";
-                }
-                return patterns[il].c_str();
-            }
-            case LAYER_FRACTION_MOE: {
-                static std::array<std::string, n_strings> patterns;
-                if (patterns[il].empty()) {
-                    patterns[il] = "blk\\." + std::to_string(il) + "\\.ffn_(up|down|gate_up|gate)_(ch|)exps";
-                }
-                return patterns[il].c_str();
-            }
-            default:
-                GGML_ABORT("fatal error");
-        }
-    };
-
-    struct ngl_t {
-        uint32_t n_layer = 0; // number of total layers
-        uint32_t n_part  = 0; // number of partial layers, <= n_layer
-
-        // for the first partial layer varying parts can overflow, all further layers use LAYER_FRACTION_MOE:
-        common_layer_fraction_t overflow_type = LAYER_FRACTION_MOE;
-
-        uint32_t n_full() const {
-            assert(n_layer >= n_part);
-            return n_layer - n_part;
-        }
-    };
-
-    const size_t ntbo = llama_max_tensor_buft_overrides();
-
-    // utility function to set n_gpu_layers and tensor_split
-    auto set_ngl_tensor_split_tbo = [&](
-            const std::vector<ngl_t> & ngl_per_device,
-            const std::vector<ggml_backend_buffer_type_t> & overflow_bufts,
-            llama_model_params & mparams) {
-        mparams.n_gpu_layers = 0;
-        for (size_t id = 0; id < nd; id++) {
-            mparams.n_gpu_layers += ngl_per_device[id].n_layer;
-            if (nd > 1) {
-                tensor_split[id] = ngl_per_device[id].n_layer;
-            }
-        }
-        assert(uint32_t(mparams.n_gpu_layers) <= hp_ngl + 1);
-        uint32_t il0 = hp_ngl + 1 - mparams.n_gpu_layers; // start index for tensor buft overrides
-
-        mparams.tensor_split = tensor_split;
-
-        size_t itbo = 0;
-        for (size_t id = 0; id < nd; id++) {
-            il0 += ngl_per_device[id].n_full();
-            for (uint32_t il = il0; il < il0 + ngl_per_device[id].n_part; il++) {
-                if (itbo + 1 >= ntbo) {
-                    tensor_buft_overrides[itbo].pattern = nullptr;
-                    tensor_buft_overrides[itbo].buft    = nullptr;
-                    itbo++;
-                    mparams.tensor_buft_overrides = tensor_buft_overrides;
-                    throw common_params_fit_exception("llama_max_tensor_buft_overrides() == "
-                        + std::to_string(ntbo) + " is insufficient for model");
-                }
-                tensor_buft_overrides[itbo].pattern = get_overflow_pattern(il, il == il0 ? ngl_per_device[id].overflow_type : LAYER_FRACTION_MOE);
-                tensor_buft_overrides[itbo].buft = il == il0 ? overflow_bufts[id] : ggml_backend_cpu_buffer_type();
-                itbo++;
-            }
-            il0 += ngl_per_device[id].n_part;
-        }
-        tensor_buft_overrides[itbo].pattern = nullptr;
-        tensor_buft_overrides[itbo].buft    = nullptr;
-        itbo++;
-        mparams.tensor_buft_overrides = tensor_buft_overrides;
-    };
-
-    // utility function that returns the memory use per device for given numbers of layers per device
-    auto get_memory_for_layers = [&](
-            const char * func_name,
-            const std::vector<ngl_t> & ngl_per_device,
-            const std::vector<ggml_backend_buffer_type_t> & overflow_bufts) -> std::vector<int64_t> {
-        llama_model_params mparams_copy = *mparams;
-        set_ngl_tensor_split_tbo(ngl_per_device, overflow_bufts, mparams_copy);
-
-        const dmds_t dmd_nl = common_get_device_memory_data(
-            path_model, &mparams_copy, cparams, devs, hp_ngl, hp_nct, hp_nex, log_level);
-
-        LOG_TRC("%s: memory for test allocation by device:\n", func_name);
-        for (size_t id = 0; id < nd; id++) {
-            const ngl_t & n = ngl_per_device[id];
-            LOG_TRC(
-                "%s: id=%zu, n_layer=%2" PRIu32 ", n_part=%2" PRIu32 ", overflow_type=%d, mem=%6" PRId64 " MiB\n",
-                func_name, id, n.n_layer, n.n_part, int(n.overflow_type), dmd_nl[id].mb.total()/MiB);
-        }
-
-        std::vector<int64_t> ret;
-        ret.reserve(nd);
-        for (size_t id = 0; id < nd; id++) {
-            ret.push_back(dmd_nl[id].mb.total());
-        }
-        return ret;
-    };
-
-    int64_t global_surplus_cpu_moe = 0;
-    if (hp_nex > 0) {
-        const static std::string pattern_moe_all = "blk\\.\\d+\\.ffn_(up|down|gate_up|gate)_(ch|)exps"; // matches all MoE tensors
-        ggml_backend_buffer_type_t cpu_buft = ggml_backend_cpu_buffer_type();
-        tensor_buft_overrides[0] = {pattern_moe_all.c_str(), cpu_buft};
-        tensor_buft_overrides[1] = {nullptr, nullptr};
-        mparams->tensor_buft_overrides = tensor_buft_overrides;
-
-        LOG_TRC("%s: getting device memory data with all MoE tensors moved to system memory:\n", __func__);
-        const dmds_t dmds_cpu_moe = common_get_device_memory_data(
-            path_model, mparams, cparams, devs, hp_ngl, hp_nct, hp_nex, log_level);
-
-        for (size_t id = 0; id < nd; id++) {
-            global_surplus_cpu_moe += dmds_cpu_moe[id].free;
-            global_surplus_cpu_moe -= int64_t(dmds_cpu_moe[id].mb.total()) + margins[id];
-        }
-
-        if (global_surplus_cpu_moe > 0) {
-            LOG_TRC("%s: with only dense weights in device memory there is a total surplus of %" PRId64 " MiB\n",
-                __func__, global_surplus_cpu_moe/MiB);
-        } else {
-            LOG_TRC("%s: with only dense weights in device memory there is still a total deficit of %" PRId64 " MiB\n",
-                __func__, -global_surplus_cpu_moe/MiB);
-        }
-
-        // reset
-        tensor_buft_overrides[0] = {nullptr, nullptr};
-        mparams->tensor_buft_overrides = tensor_buft_overrides;
-    }
-
-    std::vector<int64_t> targets; // maximum acceptable memory use per device
-    targets.reserve(nd);
-    for (size_t id = 0; id < nd; id++) {
-        targets.push_back(dmds_full[id].free - margins[id]);
-        LOG_TRC("%s: id=%zu, target=%" PRId64 " MiB\n", __func__, id, targets[id]/MiB);
-    }
-
-    std::vector<ggml_backend_buffer_type_t> overflow_bufts; // which bufts the first partial layer of a device overflows to:
-    overflow_bufts.reserve(nd);
-    for (size_t id = 0; id < nd; id++) {
-        overflow_bufts.push_back(ggml_backend_cpu_buffer_type());
-    }
-
-    std::vector<ngl_t> ngl_per_device(nd);
-    std::vector<int64_t> mem = get_memory_for_layers(__func__, ngl_per_device, overflow_bufts);
-
-    // optimize the number of layers per device using the method of false position:
-    //   - ngl_per_device has 0 layers for each device, lower bound
-    //   - try a "high" configuration where a device is given all unassigned layers
-    //   - interpolate the memory use / layer between low and high linearly to get a guess where it meets our target
-    //   - check memory use of our guess, replace either the low or high bound
-    //   - once we only have a difference of a single layer, stop and return the lower bound that just barely still fits
-    //   - the last device has the output layer, which cannot be a partial layer
-    if (hp_nex == 0) {
-        LOG_TRC("%s: filling dense layers back-to-front:\n", __func__);
-    } else {
-        LOG_TRC("%s: filling dense-only layers back-to-front:\n", __func__);
-    }
-    for (int id = nd - 1; id >= 0; id--) {
-        uint32_t n_unassigned = hp_ngl + 1;
-        for (size_t jd = id + 1; jd < nd; ++jd) {
-            assert(n_unassigned >= ngl_per_device[jd].n_layer);
-            n_unassigned -= ngl_per_device[jd].n_layer;
-        }
-
-        std::vector<ngl_t> ngl_per_device_high = ngl_per_device;
-        ngl_per_device_high[id].n_layer = n_unassigned;
-        if (hp_nex > 0) {
-            ngl_per_device_high[id].n_part = size_t(id) < nd - 1 ? ngl_per_device_high[id].n_layer : ngl_per_device_high[id].n_layer - 1;
-        }
-        if (ngl_per_device_high[id].n_layer > 0) {
-            std::vector<int64_t> mem_high = get_memory_for_layers(__func__, ngl_per_device_high, overflow_bufts);
-            if (mem_high[id] > targets[id]) {
-                assert(ngl_per_device_high[id].n_layer > ngl_per_device[id].n_layer);
-                uint32_t delta = ngl_per_device_high[id].n_layer - ngl_per_device[id].n_layer;
-                LOG_TRC("%s: start filling device %" PRIu32 ", delta=%" PRIu32 "\n", __func__, id, delta);
-                while (delta > 1) {
-                    uint32_t step_size = int64_t(delta) * (targets[id] - mem[id]) / (mem_high[id] - mem[id]);
-                    step_size = std::max(step_size, uint32_t(1));
-                    step_size = std::min(step_size, delta - 1);
-
-                    std::vector<ngl_t> ngl_per_device_test = ngl_per_device;
-                    ngl_per_device_test[id].n_layer += step_size;
-                    if (hp_nex) {
-                        ngl_per_device_test[id].n_part += size_t(id) == nd - 1 && ngl_per_device_test[id].n_part == 0 ?
-                            step_size - 1 : step_size; // the first layer is the output layer which must always be full
-                    }
-                    const std::vector<int64_t> mem_test = get_memory_for_layers(__func__, ngl_per_device_test, overflow_bufts);
-
-                    if (mem_test[id] <= targets[id]) {
-                        ngl_per_device = ngl_per_device_test;
-                        mem            = mem_test;
-                        LOG_TRC("%s: set ngl_per_device[%d].n_layer=%" PRIu32 "\n", __func__, id, ngl_per_device[id].n_layer);
-                    } else {
-                        ngl_per_device_high = ngl_per_device_test;
-                        mem_high            = mem_test;
-                        LOG_TRC("%s: set ngl_per_device_high[%d].n_layer=%" PRIu32 "\n", __func__, id, ngl_per_device_high[id].n_layer);
-                    }
-                    delta = ngl_per_device_high[id].n_layer - ngl_per_device[id].n_layer;
-                }
-            } else {
-                assert(ngl_per_device_high[id].n_layer == n_unassigned);
-                ngl_per_device = ngl_per_device_high;
-                mem            = mem_high;
-                LOG_TRC("%s: set ngl_per_device[%d].n_layer=%" PRIu32 "\n", __func__, id, ngl_per_device[id].n_layer);
-            }
-        }
-
-        const int64_t projected_margin = dmds_full[id].free - mem[id];
-        LOG_TRC(
-            "%s:   - %s: %2" PRIu32 " layers, %6" PRId64 " MiB used, %6" PRId64 " MiB free\n",
-            __func__, dev_names[id].c_str(), ngl_per_device[id].n_layer, mem[id]/MiB, projected_margin/MiB);
-    }
-    if (hp_nex == 0 || global_surplus_cpu_moe <= 0) {
-        set_ngl_tensor_split_tbo(ngl_per_device, overflow_bufts, *mparams);
-        return;
-    }
-
-    // step 4: for a MoE model where all dense tensors fit,
-    //     convert the dense-only layers in the back to full layers in the front until all devices are full
-    // essentially the same procedure as for the dense-only layers except front-to-back
-    // also, try fitting at least part of one more layer to reduce waste for "small" GPUs with e.g. 24 GiB VRAM
-
-    size_t id_dense_start = nd;
-    for (int id = nd - 1; id >= 0; id--) {
-        if (ngl_per_device[id].n_layer > 0) {
-            id_dense_start = id;
-            continue;
-        }
-        break;
-    }
-    assert(id_dense_start < nd);
-
-    LOG_TRC("%s: converting dense-only layers to full layers and filling them front-to-back with overflow to next device/system memory:\n", __func__);
-    for (size_t id = 0; id <= id_dense_start && id_dense_start < nd; id++) {
-        std::vector<ngl_t> ngl_per_device_high = ngl_per_device;
-        for (size_t jd = id_dense_start; jd < nd; jd++) {
-            const uint32_t n_layer_move = jd < nd - 1 ? ngl_per_device_high[jd].n_layer : ngl_per_device_high[jd].n_layer - 1;
-            ngl_per_device_high[id].n_layer += n_layer_move;
-            ngl_per_device_high[jd].n_layer -= n_layer_move;
-            ngl_per_device_high[jd].n_part = 0;
-        }
-        size_t id_dense_start_high = nd - 1;
-        std::vector<int64_t> mem_high = get_memory_for_layers(__func__, ngl_per_device_high, overflow_bufts);
-
-        if (mem_high[id] > targets[id]) {
-            assert(ngl_per_device_high[id].n_full() >= ngl_per_device[id].n_full());
-            uint32_t delta = ngl_per_device_high[id].n_full() - ngl_per_device[id].n_full();
-            while (delta > 1) {
-                uint32_t step_size = int64_t(delta) * (targets[id] - mem[id]) / (mem_high[id] - mem[id]);
-                step_size = std::max(step_size, uint32_t(1));
-                step_size = std::min(step_size, delta - 1);
-
-                std::vector<ngl_t> ngl_per_device_test = ngl_per_device;
-                size_t id_dense_start_test = id_dense_start;
-                uint32_t n_converted_test = 0;
-                for (;id_dense_start_test < nd; id_dense_start_test++) {
-                    const uint32_t n_convert_jd = std::min(step_size - n_converted_test, ngl_per_device_test[id_dense_start_test].n_part);
-                    ngl_per_device_test[id_dense_start_test].n_layer -= n_convert_jd;
-                    ngl_per_device_test[id_dense_start_test].n_part -= n_convert_jd;
-                    ngl_per_device_test[id].n_layer += n_convert_jd;
-                    n_converted_test += n_convert_jd;
-
-                    if (ngl_per_device_test[id_dense_start_test].n_part > 0) {
-                        break;
-                    }
-                }
-                const std::vector<int64_t> mem_test = get_memory_for_layers(__func__, ngl_per_device_test, overflow_bufts);
-
-                if (mem_test[id] <= targets[id]) {
-                    ngl_per_device = ngl_per_device_test;
-                    mem            = mem_test;
-                    id_dense_start = id_dense_start_test;
-                    LOG_TRC("%s: set ngl_per_device[%zu].(n_layer, n_part)=(%" PRIu32 ", %" PRIu32 "), id_dense_start=%zu\n",
-                        __func__, id, ngl_per_device[id].n_layer, ngl_per_device[id].n_part, id_dense_start);
-                } else {
-                    ngl_per_device_high = ngl_per_device_test;
-                    mem_high            = mem_test;
-                    id_dense_start_high = id_dense_start_test;
-                    LOG_TRC("%s: set ngl_per_device_high[%zu].(n_layer, n_part)=(%" PRIu32 ", %" PRIu32 "), id_dense_start_high=%zu\n",
-                        __func__, id, ngl_per_device_high[id].n_layer, ngl_per_device_high[id].n_part, id_dense_start_high);
-                }
-                assert(ngl_per_device_high[id].n_full() >= ngl_per_device[id].n_full());
-                delta = ngl_per_device_high[id].n_full() - ngl_per_device[id].n_full();
-            }
-        } else {
-            ngl_per_device = ngl_per_device_high;
-            mem            = mem_high;
-            id_dense_start = id_dense_start_high;
-            LOG_TRC("%s: set ngl_per_device[%zu].(n_layer, n_part)=(%" PRIu32 ", %" PRIu32 "), id_dense_start=%zu\n",
-                __func__, id, ngl_per_device[id].n_layer, ngl_per_device[id].n_part, id_dense_start);
-        }
-
-        // try to fit at least part of one more layer
-        if (ngl_per_device[id_dense_start].n_layer > (id < nd - 1 ? 0 : 1)) {
-            std::vector<ngl_t> ngl_per_device_test = ngl_per_device;
-            size_t id_dense_start_test = id_dense_start;
-            ngl_per_device_test[id_dense_start_test].n_layer--;
-            ngl_per_device_test[id_dense_start_test].n_part--;
-            ngl_per_device_test[id].n_layer++;
-            ngl_per_device_test[id].n_part++;
-            if (ngl_per_device_test[id_dense_start_test].n_part == 0) {
-                id_dense_start_test++;
-            }
-            ngl_per_device_test[id].overflow_type = LAYER_FRACTION_UP;
-            std::vector<ggml_backend_buffer_type_t> overflow_bufts_test = overflow_bufts;
-            if (id < nd - 1) {
-                overflow_bufts_test[id] = ggml_backend_dev_buffer_type(devs[id + 1]);
-            }
-            LOG_TRC("%s: trying to fit one extra layer with overflow_type=LAYER_FRACTION_UP\n", __func__);
-            std::vector<int64_t> mem_test = get_memory_for_layers(__func__, ngl_per_device_test, overflow_bufts_test);
-            if (mem_test[id] < targets[id] && (id + 1 == nd || mem_test[id + 1] < targets[id + 1])) {
-                ngl_per_device = ngl_per_device_test;
-                overflow_bufts = overflow_bufts_test;
-                mem            = mem_test;
-                id_dense_start = id_dense_start_test;
-                LOG_TRC("%s: set ngl_per_device[%zu].(n_layer, n_part, overflow_type)=(%" PRIu32 ", %" PRIu32 ", UP), id_dense_start=%zu\n",
-                    __func__, id, ngl_per_device[id].n_layer, ngl_per_device[id].n_part, id_dense_start);
-
-                ngl_per_device_test[id].overflow_type = LAYER_FRACTION_GATE;
-                LOG_TRC("%s: trying to fit one extra layer with overflow_type=LAYER_FRACTION_GATE\n", __func__);
-                mem_test = get_memory_for_layers(__func__, ngl_per_device_test, overflow_bufts_test);
-                if (mem_test[id] < targets[id] && (id + 1 == nd || mem_test[id + 1] < targets[id + 1])) {
-                    ngl_per_device = ngl_per_device_test;
-                    overflow_bufts = overflow_bufts_test;
-                    mem            = mem_test;
-                    id_dense_start = id_dense_start_test;
-                    LOG_TRC("%s: set ngl_per_device[%zu].(n_layer, n_part, overflow_type)=(%" PRIu32 ", %" PRIu32 ", GATE), id_dense_start=%zu\n",
-                        __func__, id, ngl_per_device[id].n_layer, ngl_per_device[id].n_part, id_dense_start);
-                }
-            } else {
-                ngl_per_device_test[id].overflow_type = LAYER_FRACTION_ATTN;
-                LOG_TRC("%s: trying to fit one extra layer with overflow_type=LAYER_FRACTION_ATTN\n", __func__);
-                mem_test = get_memory_for_layers(__func__, ngl_per_device_test, overflow_bufts_test);
-                if (mem_test[id] < targets[id] && (id + 1 == nd || mem_test[id + 1] < targets[id + 1])) {
-                    ngl_per_device = ngl_per_device_test;
-                    overflow_bufts = overflow_bufts_test;
-                    mem            = mem_test;
-                    id_dense_start = id_dense_start_test;
-                    LOG_TRC("%s: set ngl_per_device[%zu].(n_layer, n_part, overflow_type)=(%" PRIu32 ", %" PRIu32 ", ATTN), id_dense_start=%zu\n",
-                        __func__, id, ngl_per_device[id].n_layer, ngl_per_device[id].n_part, id_dense_start);
-                }
-            }
-        }
-
-        const int64_t projected_margin = dmds_full[id].free - mem[id];
-        LOG_TRC(
-            "%s:   - %s: %2" PRIu32 " layers (%2" PRIu32 " overflowing), %6" PRId64 " MiB used, %6" PRId64 " MiB free\n",
-            __func__, dev_names[id].c_str(), ngl_per_device[id].n_layer, ngl_per_device[id].n_part, mem[id]/MiB, projected_margin/MiB);
-    }
-
-    // print info for devices that were not changed during the conversion from dense only to full layers:
-    for (size_t id = id_dense_start + 1; id < nd; id++) {
-        const int64_t projected_margin = dmds_full[id].free - mem[id];
-        LOG_TRC(
-            "%s:   - %s: %2" PRIu32 " layers (%2" PRIu32 " overflowing), %6" PRId64 " MiB used, %6" PRId64 " MiB free\n",
-            __func__, dev_names[id].c_str(), ngl_per_device[id].n_layer, ngl_per_device[id].n_part, mem[id]/MiB, projected_margin/MiB);
-    }
-
-    set_ngl_tensor_split_tbo(ngl_per_device, overflow_bufts, *mparams);
-}
-
-enum common_params_fit_status common_fit_params(
-        const char * path_model,
-        llama_model_params * mparams,
-        llama_context_params * cparams,
-        float * tensor_split,
-        llama_model_tensor_buft_override * tensor_buft_overrides,
-        size_t * margins,
-        uint32_t n_ctx_min,
-        ggml_log_level log_level) {
-    const int64_t t0_us = llama_time_us();
-    common_params_fit_status status = COMMON_PARAMS_FIT_STATUS_SUCCESS;
-    try {
-        common_params_fit_impl(path_model, mparams, cparams, tensor_split, tensor_buft_overrides, margins, n_ctx_min, log_level);
-        LOG_TRC("%s: successfully fit params to free device memory\n", __func__);
-    } catch (const common_params_fit_exception & e) {
-        LOG_WRN("%s: failed to fit params to free device memory: %s\n", __func__, e.what());
-        status = COMMON_PARAMS_FIT_STATUS_FAILURE;
-    } catch (const std::runtime_error & e) {
-        LOG_ERR("%s: encountered an error while trying to fit params to free device memory: %s\n", __func__, e.what());
-        status = COMMON_PARAMS_FIT_STATUS_ERROR;
-    }
-    const int64_t t1_us = llama_time_us();
-    LOG_TRC("%s: fitting params to free memory took %.2f seconds\n", __func__, (t1_us - t0_us) * 1e-6);
-    return status;
-}
-
-void common_memory_breakdown_print(const struct llama_context * ctx) {
-    //const auto & devices = ctx->get_model().devices;
-    const auto * model = llama_get_model(ctx);
-
-    std::vector<ggml_backend_dev_t> devices;
-    for (int i = 0; i < llama_model_n_devices(model); i++) {
-        devices.push_back(llama_model_get_device(model, i));
-    }
-
-    llama_memory_breakdown memory_breakdown = llama_get_memory_breakdown(ctx);
-
-    std::vector<std::array<std::string, 9>> table_data;
-    table_data.reserve(devices.size());
-    const std::string template_header = "%s: | %s | %s   %s    %s   %s   %s   %s    %s |\n";
-    const std::string template_gpu    = "%s: | %s | %s = %s + (%s = %s + %s + %s) + %s |\n";
-    const std::string template_other  = "%s: | %s | %s   %s    %s = %s + %s + %s    %s |\n";
-
-    table_data.push_back({template_header, "memory breakdown [MiB]", "total", "free", "self", "model", "context", "compute", "unaccounted"});
-
-    constexpr size_t MiB = 1024 * 1024;
-    const std::vector<std::string> desc_prefixes_strip = {"NVIDIA ", "GeForce ", "Tesla ", "AMD ", "Radeon ", "Instinct "};
-
-    // track seen buffer types to avoid double counting:
-    std::set<ggml_backend_buffer_type_t> seen_buffer_types;
-
-    // accumulative memory breakdown for each device and for host:
-    std::vector<llama_memory_breakdown_data> mb_dev(devices.size());
-    llama_memory_breakdown_data              mb_host;
-
-    for (const auto & buft_mb : memory_breakdown) {
-        ggml_backend_buffer_type_t          buft = buft_mb.first;
-        const llama_memory_breakdown_data & mb   = buft_mb.second;
-        if (ggml_backend_buft_is_host(buft)) {
-            mb_host.model   += mb.model;
-            mb_host.context += mb.context;
-            mb_host.compute += mb.compute;
-            seen_buffer_types.insert(buft);
-            continue;
-        }
-        ggml_backend_dev_t dev = ggml_backend_buft_get_device(buft);
-        if (dev) {
-            int i_dev = -1;
-            for (size_t i = 0; i < devices.size(); i++) {
-                if (devices[i] == dev) {
-                    i_dev = i;
-                    break;
-                }
-            }
-            if (i_dev != -1) {
-                mb_dev[i_dev].model   += mb.model;
-                mb_dev[i_dev].context += mb.context;
-                mb_dev[i_dev].compute += mb.compute;
-                seen_buffer_types.insert(buft);
-                continue;
-            }
-        }
-    }
-
-    // print memory breakdown for each device:
-    for (size_t i = 0; i < devices.size(); i++) {
-        ggml_backend_dev_t dev = devices[i];
-        llama_memory_breakdown_data mb = mb_dev[i];
-
-        const std::string name = ggml_backend_dev_name(dev);
-        std::string desc = ggml_backend_dev_description(dev);
-        for (const std::string & prefix : desc_prefixes_strip) {
-            if (desc.length() >= prefix.length() && desc.substr(0, prefix.length()) == prefix) {
-                desc = desc.substr(prefix.length());
-            }
-        }
-
-        size_t free, total;
-        ggml_backend_dev_memory(dev, &free, &total);
-
-        const size_t self = mb.model + mb.context + mb.compute;
-        const int64_t unaccounted = static_cast<int64_t>(total) - static_cast<int64_t>(free) - static_cast<int64_t>(self);
-
-        table_data.push_back({
-            template_gpu,
-            "  - " + name + " (" + desc + ")",
-            std::to_string(total / MiB),
-            std::to_string(free / MiB),
-            std::to_string(self / MiB),
-            std::to_string(mb.model / MiB),
-            std::to_string(mb.context / MiB),
-            std::to_string(mb.compute / MiB),
-            std::to_string(unaccounted / static_cast<int64_t>(MiB))});
-    }
-
-    // print memory breakdown for host:
-    {
-        const size_t self = mb_host.model + mb_host.context + mb_host.compute;
-        table_data.push_back({
-            template_other,
-            "  - Host",
-            "", // total
-            "", // free
-            std::to_string(self / MiB),
-            std::to_string(mb_host.model / MiB),
-            std::to_string(mb_host.context / MiB),
-            std::to_string(mb_host.compute / MiB),
-            ""}); // unaccounted
-    }
-
-    // print memory breakdown for all remaining buffer types:
-    for (const auto & buft_mb : memory_breakdown) {
-        ggml_backend_buffer_type_t          buft = buft_mb.first;
-        const llama_memory_breakdown_data & mb   = buft_mb.second;
-        if (seen_buffer_types.count(buft) == 1) {
-            continue;
-        }
-        const std::string name = ggml_backend_buft_name(buft);
-        const size_t self = mb.model + mb.context + mb.compute;
-        table_data.push_back({
-            template_other,
-            "  - " + name,
-            "", // total
-            "", // free
-            std::to_string(self / MiB),
-            std::to_string(mb.model / MiB),
-            std::to_string(mb.context / MiB),
-            std::to_string(mb.compute / MiB),
-            ""}); // unaccounted
-        seen_buffer_types.insert(buft);
-    }
-
-    for (size_t j = 1; j < table_data[0].size(); j++) {
-        size_t max_len = 0;
-        for (const auto & td : table_data) {
-            max_len = std::max(max_len, td[j].length());
-        }
-        for (auto & td : table_data) {
-            td[j].insert(j == 1 ? td[j].length() : 0, max_len - td[j].length(), ' ');
-        }
-    }
-    for (const auto & td : table_data) {
-        LOG_TRC(td[0].c_str(),
-            __func__, td[1].c_str(), td[2].c_str(), td[3].c_str(), td[4].c_str(), td[5].c_str(),
-            td[6].c_str(), td[7].c_str(), td[8].c_str());
-    }
-}
-
-void common_fit_print(
-        const char * path_model,
-        llama_model_params * mparams,
-        llama_context_params * cparams) {
-    std::vector<ggml_backend_dev_t> devs;
-    uint32_t hp_ngl = 0; // hparams.n_gpu_layers
-    uint32_t hp_nct = 0; // hparams.n_ctx_train
-    uint32_t hp_nex = 0; // hparams.n_expert
-
-    auto dmd = common_get_device_memory_data(path_model, mparams, cparams, devs, hp_ngl, hp_nct, hp_nex, GGML_LOG_LEVEL_ERROR);
-    GGML_ASSERT(dmd.size() == devs.size() + 1);
-
-    for (size_t id = 0; id < devs.size(); id++) {
-        printf("%s ",  ggml_backend_dev_name(devs[id]));
-        printf("%zu ", dmd[id].mb.model/1024/1024);
-        printf("%zu ", dmd[id].mb.context/1024/1024);
-        printf("%zu ", dmd[id].mb.compute/1024/1024);
-        printf("\n");
-    }
-
-    printf("Host ");
-    printf("%zu ", dmd.back().mb.model/1024/1024);
-    printf("%zu ", dmd.back().mb.context/1024/1024);
-    printf("%zu ", dmd.back().mb.compute/1024/1024);
-    printf("\n");
-}