Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 09:15:18 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	CMakeLists.txt
#	Makefile
#	Package.swift
#	build.zig
#	tests/test-backend-ops.cpp
This commit is contained in:
Concedo
2024-04-17 18:22:47 +08:00
parent b0d796fb49 facb8b56f8
commit bfbaf0011c
16 changed files with 2005 additions and 141 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Makefile
+34 -23
View File
@@ -39,8 +39,8 @@ endif
#
# keep standard at C11 and C++11
CFLAGS = -I. -I./include -I./include/CL -I./otherarch -I./otherarch/tools -I./otherarch/sdcpp -I./otherarch/sdcpp/thirdparty -I./include/vulkan -O3 -DNDEBUG -std=c11 -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE
CXXFLAGS = -I. -I./common -I./include -I./include/CL -I./otherarch -I./otherarch/tools -I./otherarch/sdcpp -I./otherarch/sdcpp/thirdparty -I./include/vulkan -O3 -DNDEBUG -std=c++11 -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE
CFLAGS = -I. -I./include -I./include/CL -I./otherarch -I./otherarch/tools -I./otherarch/sdcpp -I./otherarch/sdcpp/thirdparty -I./include/vulkan -O3 -DNDEBUG -std=c11 -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE -DGGML_USE_LLAMAFILE
CXXFLAGS = -I. -I./common -I./include -I./include/CL -I./otherarch -I./otherarch/tools -I./otherarch/sdcpp -I./otherarch/sdcpp/thirdparty -I./include/vulkan -O3 -DNDEBUG -std=c++11 -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE -DGGML_USE_LLAMAFILE
LDFLAGS =
FASTCFLAGS = $(subst -O3,-Ofast,$(CFLAGS))
FASTCXXFLAGS = $(subst -O3,-Ofast,$(CXXFLAGS))
@@ -62,6 +62,10 @@ endif
CUBLASLD_FLAGS =
CUBLAS_OBJS =
OBJS_FULL += ggml-alloc.o ggml-backend.o ggml-quants.o unicode.o unicode-data.o sgemm.o llava.o llavaclip.o common.o grammar-parser.o
OBJS_SIMPLE += ggml-alloc.o ggml-backend.o ggml-quants_noavx2.o unicode.o unicode-data.o sgemm_noavx2.o llava.o llavaclip.o common.o grammar-parser.o
OBJS_FAILSAFE += ggml-alloc.o ggml-backend.o ggml-quants_failsafe.o unicode.o unicode-data.o sgemm_failsafe.o llava.o llavaclip.o common.o grammar-parser.o
#lets try enabling everything
CFLAGS += -pthread -s -Wno-deprecated -Wno-deprecated-declarations
CXXFLAGS += -pthread -s -Wno-multichar -Wno-write-strings -Wno-deprecated -Wno-deprecated-declarations
@@ -406,6 +410,13 @@ ggml-quants_noavx2.o: ggml-quants.c ggml.h ggml-quants.h ggml-cuda.h ggml-common
ggml-quants_failsafe.o: ggml-quants.c ggml.h ggml-quants.h ggml-cuda.h ggml-common.h
$(CC) $(CFLAGS) $(NONECFLAGS) -c $< -o $@
#sgemm
sgemm.o: sgemm.cpp sgemm.h ggml.h
$(CXX) $(CXXFLAGS) $(FULLCFLAGS) -c $< -o $@
sgemm_noavx2.o: sgemm.cpp sgemm.h ggml.h
$(CXX) $(CXXFLAGS) $(SIMPLECFLAGS) -c $< -o $@
sgemm_failsafe.o: sgemm.cpp sgemm.h ggml.h
$(CXX) $(CXXFLAGS) $(NONECFLAGS) -c $< -o $@
#there's no intrinsics or special gpu ops used here, so we can have a universal object
ggml-alloc.o: ggml-alloc.c ggml.h ggml-alloc.h
@@ -513,25 +524,25 @@ clean:
rm -vrf ggml-cuda/*.o
# useful tools
main: examples/main/main.cpp common/sampling.cpp build-info.h ggml.o ggml-quants.o ggml-alloc.o unicode.o unicode-data.o ggml-backend.o llama.o common.o console.o grammar-parser.o $(OBJS)
main: examples/main/main.cpp common/sampling.cpp build-info.h ggml.o llama.o console.o $(OBJS_FULL) $(OBJS)
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
@echo '==== Run ./main -h for help. ===='
sdmain: otherarch/sdcpp/util.cpp otherarch/sdcpp/main.cpp otherarch/sdcpp/stable-diffusion.cpp otherarch/sdcpp/upscaler.cpp otherarch/sdcpp/model.cpp otherarch/sdcpp/thirdparty/zip.c build-info.h ggml.o ggml-quants.o ggml-alloc.o unicode.o unicode-data.o ggml-backend.o llama.o common.o console.o grammar-parser.o $(OBJS)
sdmain: otherarch/sdcpp/util.cpp otherarch/sdcpp/main.cpp otherarch/sdcpp/stable-diffusion.cpp otherarch/sdcpp/upscaler.cpp otherarch/sdcpp/model.cpp otherarch/sdcpp/thirdparty/zip.c build-info.h ggml.o llama.o console.o $(OBJS_FULL) $(OBJS)
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
imatrix: examples/imatrix/imatrix.cpp common/sampling.cpp build-info.h ggml.o ggml-quants.o ggml-alloc.o unicode.o unicode-data.o ggml-backend.o llama.o common.o console.o grammar-parser.o $(OBJS)
imatrix: examples/imatrix/imatrix.cpp common/sampling.cpp build-info.h ggml.o llama.o console.o $(OBJS_FULL) $(OBJS)
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
gguf: examples/gguf/gguf.cpp build-info.h ggml.o llama.o unicode.o unicode-data.o $(OBJS)
gguf: examples/gguf/gguf.cpp build-info.h ggml.o llama.o $(OBJS_FULL) $(OBJS)
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
gguf-split: examples/gguf-split/gguf-split.cpp ggml.o ggml-quants.o ggml-alloc.o unicode.o unicode-data.o ggml-backend.o llama.o common/sampling.cpp build-info.h common.o grammar-parser.o $(OBJS)
gguf-split: examples/gguf-split/gguf-split.cpp ggml.o llama.o common/sampling.cpp build-info.h $(OBJS_FULL) $(OBJS)
$(CXX) $(CXXFLAGS) $(filter-out %.h,$^) -o $@ $(LDFLAGS)
#generated libraries
koboldcpp_default: ggml.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o common.o gpttype_adapter.o ggml-quants.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_default.o $(OBJS)
koboldcpp_default: ggml.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o gpttype_adapter.o sdcpp_default.o $(OBJS_FULL) $(OBJS)
$(DEFAULT_BUILD)
ifdef OPENBLAS_BUILD
koboldcpp_openblas: ggml_v4_openblas.o ggml_v3_openblas.o ggml_v2_openblas.o ggml_v1.o expose.o common.o gpttype_adapter.o ggml-quants.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_default.o $(OBJS)
koboldcpp_openblas: ggml_v4_openblas.o ggml_v3_openblas.o ggml_v2_openblas.o ggml_v1.o expose.o gpttype_adapter.o sdcpp_default.o $(OBJS_FULL) $(OBJS)
$(OPENBLAS_BUILD)
else
koboldcpp_openblas:
@@ -539,7 +550,7 @@ koboldcpp_openblas:
endif
ifdef FAILSAFE_BUILD
koboldcpp_failsafe: ggml_v4_failsafe.o ggml_v3_failsafe.o ggml_v2_failsafe.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_failsafe.o ggml-quants_failsafe.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_default.o $(OBJS)
koboldcpp_failsafe: ggml_v4_failsafe.o ggml_v3_failsafe.o ggml_v2_failsafe.o ggml_v1_failsafe.o expose.o gpttype_adapter_failsafe.o sdcpp_default.o $(OBJS_FAILSAFE) $(OBJS)
$(FAILSAFE_BUILD)
else
koboldcpp_failsafe:
@@ -547,7 +558,7 @@ koboldcpp_failsafe:
endif
ifdef NOAVX2_BUILD
koboldcpp_noavx2: ggml_v4_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_failsafe.o ggml-quants_noavx2.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_default.o $(OBJS)
koboldcpp_noavx2: ggml_v4_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_failsafe.o sdcpp_default.o $(OBJS_SIMPLE) $(OBJS)
$(NOAVX2_BUILD)
else
koboldcpp_noavx2:
@@ -555,10 +566,10 @@ koboldcpp_noavx2:
endif
ifdef CLBLAST_BUILD
koboldcpp_clblast: ggml_v4_clblast.o ggml_v3_clblast.o ggml_v2_clblast.o ggml_v1.o expose.o common.o gpttype_adapter_clblast.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o ggml-quants.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_default.o $(OBJS)
koboldcpp_clblast: ggml_v4_clblast.o ggml_v3_clblast.o ggml_v2_clblast.o ggml_v1.o expose.o gpttype_adapter_clblast.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o sdcpp_default.o $(OBJS_FULL) $(OBJS)
$(CLBLAST_BUILD)
ifdef NOAVX2_BUILD
koboldcpp_clblast_noavx2: ggml_v4_clblast_noavx2.o ggml_v3_clblast_noavx2.o ggml_v2_clblast_noavx2.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_clblast_noavx2.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o ggml-quants_noavx2.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_default.o $(OBJS)
koboldcpp_clblast_noavx2: ggml_v4_clblast_noavx2.o ggml_v3_clblast_noavx2.o ggml_v2_clblast_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_clblast_noavx2.o ggml-opencl.o ggml_v3-opencl.o ggml_v2-opencl.o ggml_v2-opencl-legacy.o sdcpp_default.o $(OBJS_SIMPLE) $(OBJS)
$(CLBLAST_BUILD)
else
koboldcpp_clblast_noavx2:
@@ -572,7 +583,7 @@ koboldcpp_clblast_noavx2:
endif
ifdef CUBLAS_BUILD
koboldcpp_cublas: ggml_v4_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o common.o gpttype_adapter_cublas.o ggml-quants.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_cublas.o $(CUBLAS_OBJS) $(OBJS)
koboldcpp_cublas: ggml_v4_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o gpttype_adapter_cublas.o sdcpp_cublas.o $(CUBLAS_OBJS) $(OBJS_FULL) $(OBJS)
$(CUBLAS_BUILD)
else
koboldcpp_cublas:
@@ -580,7 +591,7 @@ koboldcpp_cublas:
endif
ifdef HIPBLAS_BUILD
koboldcpp_hipblas: ggml_v4_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o common.o gpttype_adapter_cublas.o ggml-quants.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_cublas.o $(HIP_OBJS) $(OBJS)
koboldcpp_hipblas: ggml_v4_cublas.o ggml_v3_cublas.o ggml_v2_cublas.o ggml_v1.o expose.o gpttype_adapter_cublas.o sdcpp_cublas.o $(HIP_OBJS) $(OBJS_FULL) $(OBJS)
$(HIPBLAS_BUILD)
else
koboldcpp_hipblas:
@@ -588,10 +599,10 @@ koboldcpp_hipblas:
endif
ifdef VULKAN_BUILD
koboldcpp_vulkan: ggml_v4_vulkan.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o common.o gpttype_adapter_vulkan.o ggml-vulkan.o ggml-quants.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_default.o $(OBJS)
koboldcpp_vulkan: ggml_v4_vulkan.o ggml_v3.o ggml_v2.o ggml_v1.o expose.o gpttype_adapter_vulkan.o ggml-vulkan.o sdcpp_default.o $(OBJS_FULL) $(OBJS)
$(VULKAN_BUILD)
ifdef NOAVX2_BUILD
koboldcpp_vulkan_noavx2: ggml_v4_vulkan_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o common.o gpttype_adapter_vulkan_noavx2.o ggml-vulkan.o ggml-quants_noavx2.o ggml-alloc.o ggml-backend.o llava.o llavaclip.o unicode.o unicode-data.o grammar-parser.o sdcpp_default.o $(OBJS)
koboldcpp_vulkan_noavx2: ggml_v4_vulkan_noavx2.o ggml_v3_noavx2.o ggml_v2_noavx2.o ggml_v1_failsafe.o expose.o gpttype_adapter_vulkan_noavx2.o ggml-vulkan.o sdcpp_default.o $(OBJS_SIMPLE) $(OBJS)
$(VULKAN_BUILD)
else
koboldcpp_vulkan_noavx2:
@@ -605,17 +616,17 @@ koboldcpp_vulkan_noavx2:
endif
# tools
quantize_gguf: examples/quantize/quantize.cpp ggml.o llama.o ggml-quants.o ggml-alloc.o ggml-backend.o unicode.o unicode-data.o
quantize_gguf: examples/quantize/quantize.cpp ggml.o llama.o $(OBJS_FULL)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
quantize_gptj: ggml_v3.o ggml.o llama.o ggml-quants.o ggml-alloc.o ggml-backend.o unicode.o unicode-data.o otherarch/tools/gptj_quantize.cpp otherarch/tools/common-ggml.cpp
quantize_gptj: otherarch/tools/gptj_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o llama.o $(OBJS_FULL)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
quantize_gpt2: ggml_v3.o ggml.o llama.o ggml-quants.o ggml-alloc.o ggml-backend.o unicode.o unicode-data.o otherarch/tools/gpt2_quantize.cpp otherarch/tools/common-ggml.cpp
quantize_gpt2: otherarch/tools/gpt2_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o llama.o $(OBJS_FULL)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
quantize_neox: ggml_v3.o ggml.o llama.o ggml-quants.o ggml-alloc.o ggml-backend.o unicode.o unicode-data.o otherarch/tools/neox_quantize.cpp otherarch/tools/common-ggml.cpp
quantize_neox: otherarch/tools/neox_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o llama.o $(OBJS_FULL)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
quantize_mpt: ggml_v3.o ggml.o llama.o ggml-quants.o ggml-alloc.o ggml-backend.o unicode.o unicode-data.o otherarch/tools/mpt_quantize.cpp otherarch/tools/common-ggml.cpp
quantize_mpt: otherarch/tools/mpt_quantize.cpp otherarch/tools/common-ggml.cpp ggml_v3.o ggml.o llama.o $(OBJS_FULL)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
quantize_clip: ggml_v3.o ggml.o llama.o ggml-quants.o ggml-alloc.o ggml-backend.o unicode.o unicode-data.o examples/llava/clip.cpp examples/llava/clip.h examples/llava/quantclip.cpp
quantize_clip: examples/llava/clip.cpp examples/llava/clip.h examples/llava/quantclip.cpp ggml_v3.o ggml.o llama.o $(OBJS_FULL)
$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
#window simple clinfo