ci : change python3 -> python

ggml-ci

ci/run.sh

@@ -299,7 +299,7 @@ function gg_run_open_llama_7b_v2 {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python3 ../examples/convert_legacy_llama.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python ../examples/convert_legacy_llama.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
     model_q8_0="${path_models}/ggml-model-q8_0.gguf"
@@ -433,7 +433,7 @@ function gg_run_pythia_1_4b {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
     model_q8_0="${path_models}/ggml-model-q8_0.gguf"
@@ -564,7 +564,7 @@ function gg_run_pythia_2_8b {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
     model_q8_0="${path_models}/ggml-model-q8_0.gguf"
@@ -699,7 +699,7 @@ function gg_run_embd_bge_small {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
     model_q8_0="${path_models}/ggml-model-q8_0.gguf"
@@ -747,7 +747,7 @@ function gg_run_rerank_tiny {
     (time cmake -DCMAKE_BUILD_TYPE=Release ${CMAKE_EXTRA} .. ) 2>&1 | tee -a $OUT/${ci}-cmake.log
     (time make -j$(nproc)                                    ) 2>&1 | tee -a $OUT/${ci}-make.log
 
-    python3 ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
+    python ../convert_hf_to_gguf.py ${path_models} --outfile ${path_models}/ggml-model-f16.gguf
 
     model_f16="${path_models}/ggml-model-f16.gguf"
 
@@ -814,8 +814,8 @@ if [ -z ${GG_BUILD_LOW_PERF} ]; then
     mkdir -p ${mnt_models}
     ln -sfn ${mnt_models} ${SRC}/models-mnt
 
-    # Create a fresh python3 venv and enter it
-    if ! python3 -m venv "$MNT/venv"; then
+    # Create a fresh python venv and enter it
+    if ! python -m venv "$MNT/venv"; then
         echo "Error: Failed to create Python virtual environment at $MNT/venv."
         exit 1
     fi

examples/tts/README.md

@@ -78,3 +78,40 @@ play the audio:
 $ aplay output.wav
 ```
 
+### Running the example with llama-server
+Running this example with `llama-server` is also possible and requires two
+server instances to be started. One will serve the LLM model and the other
+will serve the voice decoder model.
+
+The LLM model server can be started with the following command:
+```console
+$ ./build/bin/llama-server -m ./models/outetts-0.2-0.5B-q8_0.gguf --port 8020
+```
+
+And the voice decoder model server can be started using:
+```console
+./build/bin/llama-server -m ./models/wavtokenizer-large-75-f16.gguf --port 8021 --embeddings --pooling none
+```
+
+Then we can run [tts-outetts.py](tts-outetts.py) to generate the audio.
+
+First create a virtual environment for python and install the required
+dependencies (this in only required to be done once):
+```console
+$ python3 -m venv venv
+$ source venv/bin/activate
+(venv) pip install requests numpy
+```
+
+And then run the python script using:
+```conole
+(venv) python ./examples/tts/tts-outetts.py http://localhost:8020 http://localhost:8021 "Hello world"
+spectrogram generated: n_codes: 90, n_embd: 1282
+converting to audio ...
+audio generated: 28800 samples
+audio written to file "output.wav"
+```
+And to play the audio we can again use aplay or any other media player:
+```console
+$ aplay output.wav
+```

examples/tts/tts-outetts.py

@@ -3,6 +3,121 @@ import sys
 #import struct
 import requests
 import re
+import struct
+import numpy as np
+from concurrent.futures import ThreadPoolExecutor
+
+
+def fill_hann_window(size, periodic=True):
+    if periodic:
+        return np.hanning(size + 1)[:-1]
+    return np.hanning(size)
+
+
+def irfft(n_fft, complex_input):
+    return np.fft.irfft(complex_input, n=n_fft)
+
+
+def fold(buffer, n_out, n_win, n_hop, n_pad):
+    result = np.zeros(n_out)
+    n_frames = len(buffer) // n_win
+
+    for i in range(n_frames):
+        start = i * n_hop
+        end = start + n_win
+        result[start:end] += buffer[i * n_win:(i + 1) * n_win]
+
+    return result[n_pad:-n_pad] if n_pad > 0 else result
+
+
+def process_frame(args):
+    l, n_fft, ST, hann = args
+    frame = irfft(n_fft, ST[l])
+    frame = frame * hann
+    hann2 = hann * hann
+    return frame, hann2
+
+
+def embd_to_audio(embd, n_codes, n_embd, n_thread=4):
+    embd = np.asarray(embd, dtype=np.float32).reshape(n_codes, n_embd)
+
+    n_fft = 1280
+    n_hop = 320
+    n_win = 1280
+    n_pad = (n_win - n_hop) // 2
+    n_out = (n_codes - 1) * n_hop + n_win
+
+    hann = fill_hann_window(n_fft, True)
+
+    E = np.zeros((n_embd, n_codes), dtype=np.float32)
+    for l in range(n_codes):
+        for k in range(n_embd):
+            E[k, l] = embd[l, k]
+
+    half_embd = n_embd // 2
+    S = np.zeros((n_codes, half_embd + 1), dtype=np.complex64)
+
+    for k in range(half_embd):
+        for l in range(n_codes):
+            mag = E[k, l]
+            phi = E[k + half_embd, l]
+
+            mag = np.clip(np.exp(mag), 0, 1e2)
+            S[l, k] = mag * np.exp(1j * phi)
+
+    res = np.zeros(n_codes * n_fft)
+    hann2_buffer = np.zeros(n_codes * n_fft)
+
+    with ThreadPoolExecutor(max_workers=n_thread) as executor:
+        args = [(l, n_fft, S, hann) for l in range(n_codes)]
+        results = list(executor.map(process_frame, args))
+
+        for l, (frame, hann2) in enumerate(results):
+            res[l*n_fft:(l+1)*n_fft] = frame
+            hann2_buffer[l*n_fft:(l+1)*n_fft] = hann2
+
+    audio = fold(res, n_out, n_win, n_hop, n_pad)
+    env = fold(hann2_buffer, n_out, n_win, n_hop, n_pad)
+
+    mask = env > 1e-10
+    audio[mask] /= env[mask]
+
+    return audio
+
+
+def save_wav(filename, audio_data, sample_rate):
+    num_channels = 1
+    bits_per_sample = 16
+    bytes_per_sample = bits_per_sample // 8
+    data_size = len(audio_data) * bytes_per_sample
+    byte_rate = sample_rate * num_channels * bytes_per_sample
+    block_align = num_channels * bytes_per_sample
+    chunk_size = 36 + data_size  # 36 = size of header minus first 8 bytes
+
+    header = struct.pack(
+        '<4sI4s4sIHHIIHH4sI',
+        b'RIFF',
+        chunk_size,
+        b'WAVE',
+        b'fmt ',
+        16,                # fmt chunk size
+        1,                 # audio format (PCM)
+        num_channels,
+        sample_rate,
+        byte_rate,
+        block_align,
+        bits_per_sample,
+        b'data',
+        data_size
+    )
+
+    audio_data = np.clip(audio_data * 32767, -32768, 32767)
+    pcm_data = audio_data.astype(np.int16)
+
+    with open(filename, 'wb') as f:
+        f.write(header)
+        f.write(pcm_data.tobytes())
+
 
 def process_text(text: str):
     text = re.sub(r'\d+(\.\d+)?', lambda x: x.group(), text.lower()) # TODO this needs to be fixed
@@ -170,6 +285,15 @@ n_embd = len(embd[0])
 print('spectrogram generated: n_codes: %d, n_embd: %d' % (n_codes, n_embd))
 
 # post-process the spectrogram to convert to audio
-# TODO: see the tts.cpp:embd_to_audio() and implement it in Python
 print('converting to audio ...')
-print('TODO: see the tts.cpp:embd_to_audio() and implement it in Python')
+audio = embd_to_audio(embd, n_codes, n_embd)
+print('audio generated: %d samples' % len(audio))
+
+filename = "output.wav"
+sample_rate = 24000 # sampling rate
+
+# zero out first 0.25 seconds
+audio[:24000 // 4] = 0.0
+
+save_wav(filename, audio, sample_rate)
+print('audio written to file "%s"' % filename)

ggml/CMakeLists.txt

@@ -185,6 +185,9 @@ option(GGML_OPENCL_PROFILING                "ggml: use OpenCL profiling (increas
 option(GGML_OPENCL_EMBED_KERNELS            "ggml: embed kernels"                             ON)
 option(GGML_OPENCL_USE_ADRENO_KERNELS       "ggml: use optimized kernels for Adreno"          ON)
 
+# toolchain for vulkan-shaders-gen
+set   (GGML_VULKAN_SHADERS_GEN_TOOLCHAIN "" CACHE FILEPATH "ggml: toolchain file for vulkan-shaders-gen")
+
 # extra artifacts
 option(GGML_BUILD_TESTS    "ggml: build tests"    ${GGML_STANDALONE})
 option(GGML_BUILD_EXAMPLES "ggml: build examples" ${GGML_STANDALONE})

ggml/include/ggml.h

@@ -1500,7 +1500,7 @@ extern "C" {
 
     // rotary position embedding backward, i.e compute dx from dy
     // a - dy
-    GGML_API struct ggml_tensor * ggml_rope_back(
+    GGML_API struct ggml_tensor * ggml_rope_ext_back(
             struct ggml_context * ctx,
             struct ggml_tensor  * a, // gradients of ggml_rope result
             struct ggml_tensor  * b, // positions
@@ -1515,6 +1515,23 @@ extern "C" {
             float                 beta_fast,
             float                 beta_slow);
 
+    GGML_API struct ggml_tensor * ggml_rope_multi_back(
+            struct ggml_context * ctx,
+            struct ggml_tensor  * a,
+            struct ggml_tensor  * b,
+            struct ggml_tensor  * c,
+            int                   n_dims,
+            int                   sections[4],
+            int                   mode,
+            int                   n_ctx_orig,
+            float                 freq_base,
+            float                 freq_scale,
+            float                 ext_factor,
+            float                 attn_factor,
+            float                 beta_fast,
+            float                 beta_slow);
+
+
     // clamp
     // in-place, returns view(a)
     GGML_API struct ggml_tensor * ggml_clamp(

ggml/src/ggml-cpu/ggml-cpu.c

@@ -13668,6 +13668,7 @@ struct ggml_cplan ggml_graph_plan(
                     } break;
                 case GGML_OP_SOFT_MAX:
                 case GGML_OP_ROPE:
+                case GGML_OP_ROPE_BACK:
                     {
                         cur = ggml_type_size(GGML_TYPE_F32) * node->ne[0] * n_tasks;
                     } break;

ggml/src/ggml-cpu/ggml-cpu.cpp

@@ -403,8 +403,6 @@ static bool ggml_backend_cpu_device_supports_op(ggml_backend_dev_t dev, const st
                 op->type != GGML_TYPE_IQ1_M; // missing type_traits.from_float
         case GGML_OP_MUL_MAT:
             return src1->type == GGML_TYPE_F32 || src1->type == ggml_get_type_traits_cpu(src0->type)->vec_dot_type;
-        case GGML_OP_ROPE_BACK:
-            return op->src[2] == NULL && (op->op_params[2] & 4) == 0;
         case GGML_OP_IM2COL_BACK:
             return src0->type == GGML_TYPE_F32 && src1->type == GGML_TYPE_F32;
         case GGML_OP_OUT_PROD:

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -2141,6 +2141,9 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
         case GGML_OP_ROPE:
             ggml_cuda_op_rope(ctx, dst);
             break;
+        case GGML_OP_ROPE_BACK:
+            ggml_cuda_op_rope_back(ctx, dst);
+            break;
         case GGML_OP_IM2COL:
             ggml_cuda_op_im2col(ctx, dst);
             break;
@@ -3025,7 +3028,11 @@ static bool ggml_backend_cuda_device_supports_op(ggml_backend_dev_t dev, const g
         case GGML_OP_SOFT_MAX:
             return true;
         case GGML_OP_ROPE:
-            return ggml_is_contiguous(op->src[0]);
+        case GGML_OP_ROPE_BACK: {
+            const size_t ts = ggml_type_size(op->src[0]->type);
+            const int64_t ne0_012 = op->src[0]->ne[0] * op->src[0]->ne[1] * op->src[0]->ne[2];
+            return op->src[0]->nb[0] == ts && op->src[0]->nb[3] == ne0_012*ts;
+        }
         case GGML_OP_IM2COL:
         case GGML_OP_POOL_2D:
         case GGML_OP_SUM:
@@ -3081,6 +3088,7 @@ static int64_t get_op_batch_size(const ggml_tensor * op) {
             return op->ne[1];
         case GGML_OP_MUL_MAT_ID:
         case GGML_OP_ROPE:
+        case GGML_OP_ROPE_BACK:
             return op->ne[2];
         default:
             return ggml_nrows(op);

ggml/src/ggml-cuda/rope.cu

@@ -16,9 +16,10 @@ static __device__ float rope_yarn_ramp(const float low, const float high, const
 
 // YaRN algorithm based on LlamaYaRNScaledRotaryEmbedding.py from https://github.com/jquesnelle/yarn
 // MIT licensed. Copyright (c) 2023 Jeffrey Quesnelle and Bowen Peng.
+template<bool forward>
 static __device__ void rope_yarn(
-    float theta_extrap, float freq_scale, rope_corr_dims corr_dims, int64_t i0, float ext_factor, float mscale,
-    float * cos_theta, float * sin_theta) {
+        const float theta_extrap, const float freq_scale, const rope_corr_dims corr_dims, const int64_t i0, const float ext_factor,
+        float mscale, float & cos_theta, float & sin_theta) {
     // Get n-d rotational scaling corrected for extrapolation
     float theta_interp = freq_scale * theta_extrap;
     float theta = theta_interp;
@@ -29,24 +30,28 @@ static __device__ void rope_yarn(
         // Get n-d magnitude scaling corrected for interpolation
         mscale *= 1.0f + 0.1f * logf(1.0f / freq_scale);
     }
-    *cos_theta = cosf(theta) * mscale;
-    *sin_theta = sinf(theta) * mscale;
+    cos_theta = cosf(theta) * mscale;
+    sin_theta = sinf(theta) * mscale;
+    if (!forward) {
+        sin_theta *= -1.0f;
+    }
 }
 
-template<typename T, bool has_ff>
+template<bool forward, bool has_ff, typename T>
 static __global__ void rope_norm(
-    const T * x, T * dst, int ne0, int n_dims, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float ext_factor, float attn_factor, rope_corr_dims corr_dims, float theta_scale, const float * freq_factors) {
+        const T * __restrict__ x, T * __restrict__ dst, const int ne0, const int ne1, const int s1, const int s2, const int n_dims,
+        const int32_t * __restrict__ pos, const float freq_scale, const float ext_factor, const float attn_factor,
+        const rope_corr_dims corr_dims, const float theta_scale, const float * __restrict__ freq_factors) {
     const int i0 = 2*(blockDim.y*blockIdx.y + threadIdx.y);
 
     if (i0 >= ne0) {
         return;
     }
 
-    const int row = blockDim.x*blockIdx.x + threadIdx.x;
+    const int row_dst = blockDim.x*blockIdx.x + threadIdx.x;
 
     if (i0 >= n_dims) {
-        const int i = row*ne0 + i0;
+        const int i = row_dst*ne0 + i0;
 
         dst[i + 0] = x[i + 0];
         dst[i + 1] = x[i + 1];
@@ -54,39 +59,43 @@ static __global__ void rope_norm(
         return;
     }
 
-    const int i  = row*ne0 + i0;
-    const int i2 = row/p_delta_rows;
+    const int row_x     = row_dst % ne1;
+    const int channel_x = row_dst / ne1;
 
-    const float theta_base = pos[i2]*powf(theta_scale, i0/2.0f);
+    const int idst = row_dst*ne0 + i0;
+    const int ix   = channel_x*s2 + row_x*s1 + i0;
+
+    const float theta_base = pos[channel_x]*powf(theta_scale, i0/2.0f);
 
     const float freq_factor = has_ff ? freq_factors[i0/2] : 1.0f;
 
     float cos_theta;
     float sin_theta;
 
-    rope_yarn(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, &cos_theta, &sin_theta);
+    rope_yarn<forward>(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, cos_theta, sin_theta);
 
-    const float x0 = x[i + 0];
-    const float x1 = x[i + 1];
+    const float x0 = x[ix + 0];
+    const float x1 = x[ix + 1];
 
-    dst[i + 0] = x0*cos_theta - x1*sin_theta;
-    dst[i + 1] = x0*sin_theta + x1*cos_theta;
+    dst[idst + 0] = x0*cos_theta - x1*sin_theta;
+    dst[idst + 1] = x0*sin_theta + x1*cos_theta;
 }
 
-template<typename T, bool has_ff>
+template<bool forward, bool has_ff, typename T>
 static __global__ void rope_neox(
-    const T * x, T * dst, int ne0, int n_dims, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float ext_factor, float attn_factor, rope_corr_dims corr_dims, float theta_scale, const float * freq_factors) {
+        const T * __restrict__ x, T * __restrict__ dst, const int ne0, const int ne1, const int s1, const int s2, const int n_dims,
+        const int32_t * __restrict__ pos, const float freq_scale, const float ext_factor, const float attn_factor,
+        const rope_corr_dims corr_dims, const float theta_scale, const float * __restrict__ freq_factors) {
     const int i0 = 2*(blockDim.y*blockIdx.y + threadIdx.y);
 
     if (i0 >= ne0) {
         return;
     }
 
-    const int row = blockDim.x*blockIdx.x + threadIdx.x;
+    const int row_dst = blockDim.x*blockIdx.x + threadIdx.x;
 
     if (i0 >= n_dims) {
-        const int i = row*ne0 + i0;
+        const int i = row_dst*ne0 + i0;
 
         dst[i + 0] = x[i + 0];
         dst[i + 1] = x[i + 1];
@@ -94,39 +103,43 @@ static __global__ void rope_neox(
         return;
     }
 
-    const int i  = row*ne0 + i0/2;
-    const int i2 = row/p_delta_rows;
+    const int row_x     = row_dst % ne1;
+    const int channel_x = row_dst / ne1;
 
-    const float theta_base = pos[i2]*powf(theta_scale, i0/2.0f);
+    const int idst = row_dst*ne0 + i0/2;
+    const int ix   = channel_x*s2 + row_x*s1 + i0/2;
+
+    const float theta_base = pos[channel_x]*powf(theta_scale, i0/2.0f);
 
     const float freq_factor = has_ff ? freq_factors[i0/2] : 1.0f;
 
     float cos_theta;
     float sin_theta;
 
-    rope_yarn(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, &cos_theta, &sin_theta);
+    rope_yarn<forward>(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, cos_theta, sin_theta);
 
-    const float x0 = x[i + 0];
-    const float x1 = x[i + n_dims/2];
+    const float x0 = x[ix + 0];
+    const float x1 = x[ix + n_dims/2];
 
-    dst[i + 0]        = x0*cos_theta - x1*sin_theta;
-    dst[i + n_dims/2] = x0*sin_theta + x1*cos_theta;
+    dst[idst + 0]        = x0*cos_theta - x1*sin_theta;
+    dst[idst + n_dims/2] = x0*sin_theta + x1*cos_theta;
 }
 
-template<typename T, bool has_ff>
+template<bool forward, bool has_ff, typename T>
 static __global__ void rope_multi(
-    const T * x, T * dst, int ne0, int ne2, int n_dims, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float ext_factor, float attn_factor, rope_corr_dims corr_dims, float theta_scale, const float * freq_factors, mrope_sections sections) {
+        const T * __restrict__ x, T * __restrict__ dst, const int ne0, const int ne1, const int ne2, const int s1, const int s2,
+        const int n_dims, const int32_t * __restrict__ pos, const float freq_scale, const float ext_factor, const float attn_factor,
+        const rope_corr_dims corr_dims, const float theta_scale, const float * __restrict__ freq_factors, const mrope_sections sections) {
     const int i0 = 2*(blockDim.y*blockIdx.y + threadIdx.y);
 
     if (i0 >= ne0) {
         return;
     }
 
-    const int row = blockDim.x*blockIdx.x + threadIdx.x;
+    const int row_dst = blockDim.x*blockIdx.x + threadIdx.x;
 
     if (i0 >= n_dims) {
-        const int i = row*ne0 + i0;
+        const int i = row_dst*ne0 + i0;
 
         dst[i + 0] = x[i + 0];
         dst[i + 1] = x[i + 1];
@@ -134,25 +147,28 @@ static __global__ void rope_multi(
         return;
     }
 
-    const int i  = row*ne0 + i0/2;
-    const int i2 = row/p_delta_rows;
+    const int row_x     = row_dst % ne1;
+    const int channel_x = row_dst / ne1;
 
-    int sect_dims = sections.v[0] + sections.v[1] + sections.v[2] + sections.v[3];
-    int sec_w = sections.v[1] + sections.v[0];
-    int sector = (i0 / 2) % sect_dims;
+    const int idst = row_dst*ne0 + i0/2;
+    const int ix   = channel_x*s2 + row_x*s1 + i0/2;
+
+    const int sect_dims = sections.v[0] + sections.v[1] + sections.v[2] + sections.v[3];
+    const int sec_w = sections.v[1] + sections.v[0];
+    const int sector = (i0 / 2) % sect_dims;
 
     float theta_base = 0.0;
     if (sector < sections.v[0]) {
-        theta_base = pos[i2]*powf(theta_scale, i0/2.0f);
+        theta_base = pos[channel_x]*powf(theta_scale, i0/2.0f);
     }
     else if (sector >= sections.v[0] && sector < sec_w) {
-        theta_base = pos[i2 + ne2 * 1]*powf(theta_scale, i0/2.0f);
+        theta_base = pos[channel_x + ne2 * 1]*powf(theta_scale, i0/2.0f);
     }
     else if (sector >= sec_w && sector < sec_w + sections.v[2]) {
-        theta_base = pos[i2 + ne2 * 2]*powf(theta_scale, i0/2.0f);
+        theta_base = pos[channel_x + ne2 * 2]*powf(theta_scale, i0/2.0f);
     }
     else if (sector >= sec_w + sections.v[2]) {
-        theta_base = pos[i2 + ne2 * 3]*powf(theta_scale, i0/2.0f);
+        theta_base = pos[channel_x + ne2 * 3]*powf(theta_scale, i0/2.0f);
     }
 
     const float freq_factor = has_ff ? freq_factors[i0/2] : 1.0f;
@@ -160,42 +176,46 @@ static __global__ void rope_multi(
     float cos_theta;
     float sin_theta;
 
-    rope_yarn(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, &cos_theta, &sin_theta);
+    rope_yarn<forward>(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, cos_theta, sin_theta);
 
-    const float x0 = x[i + 0];
-    const float x1 = x[i + n_dims/2];
+    const float x0 = x[ix + 0];
+    const float x1 = x[ix + n_dims/2];
 
-    dst[i + 0]        = x0*cos_theta - x1*sin_theta;
-    dst[i + n_dims/2] = x0*sin_theta + x1*cos_theta;
+    dst[idst + 0]        = x0*cos_theta - x1*sin_theta;
+    dst[idst + n_dims/2] = x0*sin_theta + x1*cos_theta;
 }
 
-template<typename T, bool has_ff>
+template<bool forward, bool has_ff, typename T>
 static __global__ void rope_vision(
-    const T * x, T * dst, int ne0, int ne2, int n_dims, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float ext_factor, float attn_factor, rope_corr_dims corr_dims, float theta_scale, const float * freq_factors, mrope_sections sections) {
+        const T * __restrict__ x, T * __restrict__ dst, const int ne0, const int ne1, const int ne2, const int s1, const int s2, const int n_dims,
+        const int32_t * __restrict__ pos, const float freq_scale, const float ext_factor, const float attn_factor, const rope_corr_dims corr_dims,
+        const float theta_scale, const float * __restrict__ freq_factors, const mrope_sections sections) {
     const int i0 = 2*(blockDim.y*blockIdx.y + threadIdx.y);
 
     if (i0 >= ne0) {
         return;
     }
 
-    const int row = blockDim.x*blockIdx.x + threadIdx.x;
+    const int row_dst = blockDim.x*blockIdx.x + threadIdx.x;
 
-    const int i  = row*ne0 + i0/2;
-    const int i2 = row/p_delta_rows; // i2-th tokens
+    const int row_x     = row_dst % ne1;
+    const int channel_x = row_dst / ne1;
 
-    int sect_dims = sections.v[0] + sections.v[1];
-    int sec_w = sections.v[1] + sections.v[0];
-    int sector = (i0 / 2) % sect_dims;
+    const int idst = row_dst*ne0 + i0/2;
+    const int ix   = channel_x*s2 + row_x*s1 + i0/2;
+
+    const int sect_dims = sections.v[0] + sections.v[1];
+    const int sec_w = sections.v[1] + sections.v[0];
+    const int sector = (i0 / 2) % sect_dims;
 
     float theta_base = 0.0;
     if (sector < sections.v[0]) {
         const int p = sector;
-        theta_base = pos[i2]*powf(theta_scale, p);
+        theta_base = pos[channel_x]*powf(theta_scale, p);
     }
     else if (sector >= sections.v[0] && sector < sec_w) {
         const int p = sector - sections.v[0];
-        theta_base = pos[i2 + ne2]*powf(theta_scale, p);
+        theta_base = pos[channel_x + ne2]*powf(theta_scale, p);
     }
 
     const float freq_factor = has_ff ? freq_factors[i0/2] : 1.0f;
@@ -203,19 +223,20 @@ static __global__ void rope_vision(
     float cos_theta;
     float sin_theta;
 
-    rope_yarn(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, &cos_theta, &sin_theta);
+    rope_yarn<forward>(theta_base/freq_factor, freq_scale, corr_dims, i0, ext_factor, attn_factor, cos_theta, sin_theta);
 
-    const float x0 = x[i + 0];
-    const float x1 = x[i + n_dims];
+    const float x0 = x[ix + 0];
+    const float x1 = x[ix + n_dims];
 
-    dst[i + 0]      = x0*cos_theta - x1*sin_theta;
-    dst[i + n_dims] = x0*sin_theta + x1*cos_theta;
+    dst[idst + 0]      = x0*cos_theta - x1*sin_theta;
+    dst[idst + n_dims] = x0*sin_theta + x1*cos_theta;
 }
 
-template<typename T>
+template<bool forward, typename T>
 static void rope_norm_cuda(
-    const T * x, T * dst, int ne0, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, cudaStream_t stream) {
+        const T * __restrict__ x, T * __restrict__ dst, const int ne0, const int ne1, const int s1, const int s2, const int n_dims, const int nr,
+        const int32_t * __restrict__ pos, const float freq_scale, const float freq_base, const float ext_factor, const float attn_factor,
+        const rope_corr_dims corr_dims, const float * __restrict__ freq_factors, cudaStream_t stream) {
     GGML_ASSERT(ne0 % 2 == 0);
     const dim3 block_dims(1, CUDA_ROPE_BLOCK_SIZE, 1);
     const int n_blocks_x = (ne0 + 2*CUDA_ROPE_BLOCK_SIZE - 1) / (2*CUDA_ROPE_BLOCK_SIZE);
@@ -224,22 +245,21 @@ static void rope_norm_cuda(
     const float theta_scale = powf(freq_base, -2.0f/n_dims);
 
     if (freq_factors == nullptr) {
-        rope_norm<T, false><<<block_nums, block_dims, 0, stream>>>(
-                x, dst, ne0, n_dims, pos, freq_scale, p_delta_rows, ext_factor, attn_factor, corr_dims,
-                theta_scale, freq_factors
-                );
+        rope_norm<forward, false><<<block_nums, block_dims, 0, stream>>>(
+            x, dst, ne0, ne1, s1, s2, n_dims, pos, freq_scale, ext_factor,
+            attn_factor, corr_dims, theta_scale, freq_factors);
     } else {
-        rope_norm<T, true><<<block_nums, block_dims, 0, stream>>>(
-                x, dst, ne0, n_dims, pos, freq_scale, p_delta_rows, ext_factor, attn_factor, corr_dims,
-                theta_scale, freq_factors
-                );
+        rope_norm<forward, true><<<block_nums, block_dims, 0, stream>>>(
+            x, dst, ne0, ne1, s1, s2, n_dims, pos, freq_scale, ext_factor,
+            attn_factor, corr_dims, theta_scale, freq_factors);
     }
 }
 
-template<typename T>
+template<bool forward, typename T>
 static void rope_neox_cuda(
-    const T * x, T * dst, int ne0, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, cudaStream_t stream) {
+        const T * __restrict__ x, T * __restrict__ dst, const int ne0, const int ne1, const int s1, const int s2, const int n_dims, const int nr,
+        const int32_t * __restrict__ pos, const float freq_scale, const float freq_base, const float ext_factor, const float attn_factor,
+        const rope_corr_dims corr_dims, const float * __restrict__ freq_factors, cudaStream_t stream) {
     GGML_ASSERT(ne0 % 2 == 0);
     const dim3 block_dims(1, CUDA_ROPE_BLOCK_SIZE, 1);
     const int n_blocks_x = (ne0 + 2*CUDA_ROPE_BLOCK_SIZE - 1) / (2*CUDA_ROPE_BLOCK_SIZE);
@@ -248,22 +268,21 @@ static void rope_neox_cuda(
     const float theta_scale = powf(freq_base, -2.0f/n_dims);
 
     if (freq_factors == nullptr) {
-        rope_neox<T, false><<<block_nums, block_dims, 0, stream>>>(
-                x, dst, ne0, n_dims, pos, freq_scale, p_delta_rows, ext_factor, attn_factor, corr_dims,
-                theta_scale, freq_factors
-                );
+        rope_neox<forward, false, T><<<block_nums, block_dims, 0, stream>>>(
+            x, dst, ne0, ne1, s1, s2, n_dims, pos, freq_scale, ext_factor,
+            attn_factor, corr_dims, theta_scale, freq_factors);
     } else {
-        rope_neox<T, true><<<block_nums, block_dims, 0, stream>>>(
-                x, dst, ne0, n_dims, pos, freq_scale, p_delta_rows, ext_factor, attn_factor, corr_dims,
-                theta_scale, freq_factors
-                );
+        rope_neox<forward, true, T><<<block_nums, block_dims, 0, stream>>>(
+            x, dst, ne0, ne1, s1, s2, n_dims, pos, freq_scale, ext_factor,
+            attn_factor, corr_dims, theta_scale, freq_factors);
     }
 }
 
-template<typename T>
+template<bool forward, typename T>
 static void rope_multi_cuda(
-    const T * x, T * dst, int ne0, int ne2, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, mrope_sections sections, cudaStream_t stream) {
+        const T * __restrict__ x, T * __restrict__ dst, const int ne0, const int ne1, const int ne2, const int s1, const int s2, const int n_dims, const int nr,
+        const int32_t * __restrict__ pos, const float freq_scale, const float freq_base, const float ext_factor, const float attn_factor,
+        const rope_corr_dims corr_dims, const float * __restrict__ freq_factors, const mrope_sections sections, cudaStream_t stream) {
     GGML_ASSERT(ne0 % 2 == 0);
     const dim3 block_dims(1, CUDA_ROPE_BLOCK_SIZE, 1);
     const int n_blocks_x = (ne0 + 2*CUDA_ROPE_BLOCK_SIZE - 1) / (2*CUDA_ROPE_BLOCK_SIZE);
@@ -272,22 +291,21 @@ static void rope_multi_cuda(
     const float theta_scale = powf(freq_base, -2.0f/n_dims);
 
     if (freq_factors == nullptr) {
-        rope_multi<T, false><<<block_nums, block_dims, 0, stream>>>(
-                x, dst, ne0, ne2, n_dims, pos, freq_scale, p_delta_rows, ext_factor, attn_factor, corr_dims,
-                theta_scale, freq_factors, sections
-                );
+        rope_multi<forward, false, T><<<block_nums, block_dims, 0, stream>>>(
+            x, dst, ne0, ne1, ne2, s1, s2, n_dims, pos, freq_scale, ext_factor,
+            attn_factor, corr_dims, theta_scale, freq_factors, sections);
     } else {
-        rope_multi<T, true><<<block_nums, block_dims, 0, stream>>>(
-                x, dst, ne0, ne2, n_dims, pos, freq_scale, p_delta_rows, ext_factor, attn_factor, corr_dims,
-                theta_scale, freq_factors, sections
-                );
+        rope_multi<forward, true, T><<<block_nums, block_dims, 0, stream>>>(
+            x, dst, ne0, ne1, ne2, s1, s2, n_dims, pos, freq_scale, ext_factor,
+            attn_factor, corr_dims, theta_scale, freq_factors, sections);
     }
 }
 
-template<typename T>
+template<bool forward, typename T>
 static void rope_vision_cuda(
-    const T * x, T * dst, int ne0, int ne2, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, mrope_sections sections, cudaStream_t stream) {
+        const T * __restrict__ x, T * __restrict__ dst, const int ne0, const int ne1, const int ne2, const int s1, const int s2, const int n_dims, const int nr,
+        const int32_t * __restrict__ pos, const float freq_scale, const float freq_base, const float ext_factor, const float attn_factor,
+        const rope_corr_dims corr_dims, const float * __restrict__ freq_factors, const mrope_sections sections, cudaStream_t stream) {
     GGML_ASSERT(ne0 % 2 == 0);
     const dim3 block_dims(1, CUDA_ROPE_BLOCK_SIZE, 1);
     const int n_blocks_x = (ne0 + 2*CUDA_ROPE_BLOCK_SIZE - 1) / (2*CUDA_ROPE_BLOCK_SIZE);
@@ -298,80 +316,18 @@ static void rope_vision_cuda(
     const float theta_scale = powf(freq_base, -2.0f/n_dims);
 
     if (freq_factors == nullptr) {
-        rope_vision<T, false><<<block_nums, block_dims, 0, stream>>>(
-                x, dst, ne0, ne2, n_dims, pos, freq_scale, p_delta_rows, ext_factor, attn_factor, corr_dims,
-                theta_scale, freq_factors, sections
-                );
+        rope_vision<forward, false, T><<<block_nums, block_dims, 0, stream>>>(
+            x, dst, ne0, ne1, ne2, s1, s2, n_dims, pos, freq_scale, ext_factor,
+            attn_factor, corr_dims, theta_scale, freq_factors, sections);
     } else {
-        rope_vision<T, true><<<block_nums, block_dims, 0, stream>>>(
-                x, dst, ne0, ne2, n_dims, pos, freq_scale, p_delta_rows, ext_factor, attn_factor, corr_dims,
-                theta_scale, freq_factors, sections
-                );
+        rope_vision<forward, true, T><<<block_nums, block_dims, 0, stream>>>(
+            x, dst, ne0, ne1, ne2, s1, s2, n_dims, pos, freq_scale, ext_factor,
+            attn_factor, corr_dims, theta_scale, freq_factors, sections);
     }
 }
 
-static void rope_norm_cuda_f16(
-    const half * x, half * dst, int ne0, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, cudaStream_t stream) {
-
-    rope_norm_cuda<half>(x, dst, ne0, n_dims, nr, pos, freq_scale, p_delta_rows, freq_base, ext_factor, attn_factor, corr_dims, freq_factors, stream);
-}
-
-static void rope_norm_cuda_f32(
-    const float * x, float * dst, int ne0, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, cudaStream_t stream) {
-
-    rope_norm_cuda<float>(x, dst, ne0, n_dims, nr, pos, freq_scale, p_delta_rows, freq_base, ext_factor, attn_factor, corr_dims, freq_factors, stream);
-}
-
-static void rope_neox_cuda_f16(
-    const half * x, half * dst, int ne0, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, cudaStream_t stream) {
-
-    rope_neox_cuda<half>(x, dst, ne0, n_dims, nr, pos, freq_scale, p_delta_rows, freq_base, ext_factor, attn_factor, corr_dims, freq_factors, stream);
-}
-
-static void rope_neox_cuda_f32(
-    const float * x, float * dst, int ne0, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, cudaStream_t stream
-) {
-
-    rope_neox_cuda<float>(x, dst, ne0, n_dims, nr, pos, freq_scale, p_delta_rows, freq_base, ext_factor, attn_factor, corr_dims, freq_factors, stream);
-}
-
-static void rope_multi_cuda_f16(
-    const half * x, half * dst, int ne0, int ne2, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, mrope_sections sections, cudaStream_t stream
-) {
-
-    rope_multi_cuda<half>(x, dst, ne0, ne2, n_dims, nr, pos, freq_scale, p_delta_rows, freq_base, ext_factor, attn_factor, corr_dims, freq_factors, sections, stream);
-}
-
-static void rope_multi_cuda_f32(
-    const float * x, float * dst, int ne0, int ne2, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, mrope_sections sections, cudaStream_t stream
-) {
-
-    rope_multi_cuda<float>(x, dst, ne0, ne2, n_dims, nr, pos, freq_scale, p_delta_rows, freq_base, ext_factor, attn_factor, corr_dims, freq_factors, sections, stream);
-}
-
-static void rope_vision_cuda_f16(
-    const half * x, half * dst, int ne0, int ne2, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, mrope_sections sections, cudaStream_t stream
-) {
-
-    rope_vision_cuda<half>(x, dst, ne0, ne2, n_dims, nr, pos, freq_scale, p_delta_rows, freq_base, ext_factor, attn_factor, corr_dims, freq_factors, sections, stream);
-}
-
-static void rope_vision_cuda_f32(
-    const float * x, float * dst, int ne0, int ne2, int n_dims, int nr, const int32_t * pos, float freq_scale, int p_delta_rows,
-    float freq_base, float ext_factor, float attn_factor, rope_corr_dims corr_dims, const float * freq_factors, mrope_sections sections, cudaStream_t stream
-) {
-
-    rope_vision_cuda<float>(x, dst, ne0, ne2, n_dims, nr, pos, freq_scale, p_delta_rows, freq_base, ext_factor, attn_factor, corr_dims, freq_factors, sections, stream);
-}
-
-void ggml_cuda_op_rope(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+template <bool forward>
+void ggml_cuda_op_rope_impl(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
     const ggml_tensor * src0 = dst->src[0];
     const ggml_tensor * src1 = dst->src[1];
     const ggml_tensor * src2 = dst->src[2];
@@ -382,7 +338,6 @@ void ggml_cuda_op_rope(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
     float * dst_d = (float *)dst->data;
     cudaStream_t stream = ctx.stream();
 
-    GGML_ASSERT(ggml_is_contiguous(src0));
     GGML_ASSERT(src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_F16);
     GGML_ASSERT( dst->type == GGML_TYPE_F32 ||  dst->type == GGML_TYPE_F16);
     GGML_ASSERT(src0->type == dst->type);
@@ -392,6 +347,9 @@ void ggml_cuda_op_rope(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
     const int64_t ne02 = src0->ne[2]; // num heads
     const int64_t nr = ggml_nrows(src0);
 
+    const size_t s01 = src0->nb[1] / ggml_type_size(src0->type);
+    const size_t s02 = src0->nb[2] / ggml_type_size(src0->type);
+
     //const int n_past     = ((int32_t *) dst->op_params)[0];
     const int n_dims     = ((int32_t *) dst->op_params)[1];
     const int mode       = ((int32_t *) dst->op_params)[2];
@@ -440,59 +398,59 @@ void ggml_cuda_op_rope(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
     // compute
     if (is_neox) {
         if (src0->type == GGML_TYPE_F32) {
-            rope_neox_cuda_f32(
-                (const float *)src0_d, (float *)dst_d, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, stream
-            );
+            rope_neox_cuda<forward>(
+                (const float *) src0_d, (float *) dst_d, ne00, ne01, s01, s02, n_dims, nr, pos, freq_scale,
+                freq_base, ext_factor, attn_factor, corr_dims, freq_factors, stream);
         } else if (src0->type == GGML_TYPE_F16) {
-            rope_neox_cuda_f16(
-                (const half *)src0_d, (half *)dst_d, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, stream
-            );
+            rope_neox_cuda<forward>(
+                (const half *) src0_d, (half *) dst_d, ne00, ne01, s01, s02, n_dims, nr, pos, freq_scale,
+                freq_base, ext_factor, attn_factor, corr_dims, freq_factors, stream);
         } else {
             GGML_ABORT("fatal error");
         }
     } else if (is_mrope && !is_vision) {
         if (src0->type == GGML_TYPE_F32) {
-            rope_multi_cuda_f32(
-                (const float *)src0_d, (float *)dst_d, ne00, ne02, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, sections, stream
-            );
+            rope_multi_cuda<forward>(
+                (const float *) src0_d, (float *) dst_d, ne00, ne01, ne02, s01, s02, n_dims, nr, pos, freq_scale,
+                freq_base, ext_factor, attn_factor, corr_dims, freq_factors, sections, stream);
         } else if (src0->type == GGML_TYPE_F16) {
-            rope_multi_cuda_f16(
-                (const half *)src0_d, (half *)dst_d, ne00, ne02, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, sections, stream
-            );
+            rope_multi_cuda<forward>(
+                (const half *) src0_d, (half *) dst_d, ne00, ne01, ne02, s01, s02, n_dims, nr, pos, freq_scale,
+                freq_base, ext_factor, attn_factor, corr_dims, freq_factors, sections, stream);
         } else {
             GGML_ABORT("fatal error");
         }
     } else if (is_vision) {
         if (src0->type == GGML_TYPE_F32) {
-            rope_vision_cuda_f32(
-                (const float *)src0_d, (float *)dst_d, ne00, ne02, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, sections, stream
-            );
+            rope_vision_cuda<forward>(
+                (const float *) src0_d, (float *) dst_d, ne00, ne01, ne02, s01, s02, n_dims, nr, pos, freq_scale,
+                freq_base, ext_factor, attn_factor, corr_dims, freq_factors, sections, stream);
         } else if (src0->type == GGML_TYPE_F16) {
-            rope_vision_cuda_f16(
-                (const half *)src0_d, (half *)dst_d, ne00, ne02, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, sections, stream
-            );
+            rope_vision_cuda<forward>(
+                (const half *) src0_d, (half *) dst_d, ne00, ne01, ne02, s01, s02, n_dims, nr, pos, freq_scale,
+                freq_base, ext_factor, attn_factor, corr_dims, freq_factors, sections, stream);
         } else {
             GGML_ABORT("fatal error");
         }
     } else {
         if (src0->type == GGML_TYPE_F32) {
-            rope_norm_cuda_f32(
-                (const float *)src0_d, (float *)dst_d, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, stream
-            );
+            rope_norm_cuda<forward>(
+                (const float *) src0_d, (float *) dst_d, ne00, ne01, s01, s02, n_dims, nr, pos, freq_scale,
+                freq_base, ext_factor, attn_factor, corr_dims, freq_factors, stream);
         } else if (src0->type == GGML_TYPE_F16) {
-            rope_norm_cuda_f16(
-                (const half *)src0_d, (half *)dst_d, ne00, n_dims, nr, pos, freq_scale, ne01, freq_base, ext_factor,
-                attn_factor, corr_dims, freq_factors, stream
-            );
+            rope_norm_cuda<forward>(
+                (const half *) src0_d, (half *) dst_d, ne00, ne01, s01, s02, n_dims, nr, pos, freq_scale,
+                freq_base, ext_factor, attn_factor, corr_dims, freq_factors, stream);
         } else {
             GGML_ABORT("fatal error");
         }
     }
 }
+
+void ggml_cuda_op_rope(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    ggml_cuda_op_rope_impl<true>(ctx, dst);
+}
+
+void ggml_cuda_op_rope_back(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {
+    ggml_cuda_op_rope_impl<false>(ctx, dst);
+}

ggml/src/ggml-cuda/rope.cuh

@@ -3,3 +3,5 @@
 #define CUDA_ROPE_BLOCK_SIZE 256
 
 void ggml_cuda_op_rope(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
+
+void ggml_cuda_op_rope_back(ggml_backend_cuda_context & ctx, ggml_tensor * dst);

ggml/src/ggml-vulkan/CMakeLists.txt

@@ -1,5 +1,20 @@
+cmake_minimum_required(VERSION 3.19)
+cmake_policy(SET CMP0114 NEW)
+
 find_package(Vulkan COMPONENTS glslc REQUIRED)
 
+function(detect_host_compiler)
+    if (CMAKE_HOST_SYSTEM_NAME STREQUAL "Windows")
+        find_program(HOST_C_COMPILER NAMES cl gcc clang NO_CMAKE_FIND_ROOT_PATH)
+        find_program(HOST_CXX_COMPILER NAMES cl g++ clang++ NO_CMAKE_FIND_ROOT_PATH)
+    else()
+        find_program(HOST_C_COMPILER NAMES gcc clang NO_CMAKE_FIND_ROOT_PATH)
+        find_program(HOST_CXX_COMPILER NAMES g++ clang++ NO_CMAKE_FIND_ROOT_PATH)
+    endif()
+    set(HOST_C_COMPILER "${HOST_C_COMPILER}" PARENT_SCOPE)
+    set(HOST_CXX_COMPILER "${HOST_CXX_COMPILER}" PARENT_SCOPE)
+endfunction()
+
 if (Vulkan_FOUND)
     message(STATUS "Vulkan found")
 
@@ -73,19 +88,56 @@ if (Vulkan_FOUND)
         add_compile_definitions(GGML_VULKAN_RUN_TESTS)
     endif()
 
-    add_subdirectory(vulkan-shaders)
+    if (NOT CMAKE_CROSSCOMPILING)
+        add_subdirectory(vulkan-shaders)
+        if (MSVC)
+            foreach(CONFIG ${CMAKE_CONFIGURATION_TYPES})
+                string(TOUPPER ${CONFIG} CONFIG)
+                set_target_properties(vulkan-shaders-gen PROPERTIES
+                    RUNTIME_OUTPUT_DIRECTORY_${CONFIG} ${CMAKE_RUNTIME_OUTPUT_DIRECTORY})
+            endforeach()
+        endif()
+    else()
+        if (GGML_VULKAN_SHADERS_GEN_TOOLCHAIN)
+            set(HOST_CMAKE_TOOLCHAIN_FILE ${GGML_VULKAN_SHADERS_GEN_TOOLCHAIN})
+        else()
+            detect_host_compiler()
+            if (NOT HOST_C_COMPILER OR NOT HOST_CXX_COMPILER)
+                message(FATAL_ERROR "Host compiler not found")
+            else()
+                message(STATUS "Host compiler: ${HOST_C_COMPILER} ${HOST_CXX_COMPILER}")
+            endif()
+            configure_file(${CMAKE_CURRENT_SOURCE_DIR}/cmake/host-toolchain.cmake.in ${CMAKE_BINARY_DIR}/host-toolchain.cmake @ONLY)
+            set(HOST_CMAKE_TOOLCHAIN_FILE ${CMAKE_BINARY_DIR}/host-toolchain.cmake)
+        endif()
+        message(STATUS "vulkan-shaders-gen toolchain file: ${HOST_CMAKE_TOOLCHAIN_FILE}")
 
-    set (_ggml_vk_genshaders_cmd vulkan-shaders-gen)
+        include(ExternalProject)
+        # Native build through ExternalProject_Add
+        ExternalProject_Add(
+            vulkan-shaders-gen
+            SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/vulkan-shaders
+            CMAKE_ARGS -DCMAKE_TOOLCHAIN_FILE=${HOST_CMAKE_TOOLCHAIN_FILE}
+                    -DCMAKE_INSTALL_PREFIX=${CMAKE_BINARY_DIR}
+            BUILD_COMMAND ${CMAKE_COMMAND} --build .
+            INSTALL_COMMAND ${CMAKE_COMMAND} --install .
+            INSTALL_DIR ${CMAKE_BINARY_DIR}
+        )
+        ExternalProject_Add_StepTargets(vulkan-shaders-gen build install)
+    endif()
+    set (_ggml_vk_host_suffix $<IF:$<STREQUAL:${CMAKE_HOST_SYSTEM_NAME},Windows>,.exe,>)
+    set (_ggml_vk_genshaders_cmd ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/vulkan-shaders-gen${_ggml_vk_host_suffix})
     set (_ggml_vk_header     ${CMAKE_CURRENT_BINARY_DIR}/ggml-vulkan-shaders.hpp)
     set (_ggml_vk_source     ${CMAKE_CURRENT_BINARY_DIR}/ggml-vulkan-shaders.cpp)
     set (_ggml_vk_input_dir  ${CMAKE_CURRENT_SOURCE_DIR}/vulkan-shaders)
     set (_ggml_vk_output_dir ${CMAKE_CURRENT_BINARY_DIR}/vulkan-shaders.spv)
 
     file(GLOB _ggml_vk_shader_deps "${_ggml_vk_input_dir}/*.comp")
+    set (_ggml_vk_shader_deps ${_ggml_vk_shader_deps} vulkan-shaders-gen)
 
-    if (NOT CMAKE_CROSSCOMPILING)
-        set(_ggml_vk_genshaders_cmd "$<TARGET_FILE_DIR:vulkan-shaders-gen>/${_ggml_vk_genshaders_cmd}")
-    endif ()
+    if (CMAKE_CROSSCOMPILING)
+        set(_ggml_vk_shader_deps ${_ggml_vk_shader_deps} vulkan-shaders-gen-build vulkan-shaders-gen-install)
+    endif()
 
     add_custom_command(
         OUTPUT ${_ggml_vk_header}
@@ -99,7 +151,7 @@ if (Vulkan_FOUND)
             --target-cpp ${_ggml_vk_source}
             --no-clean
 
-        DEPENDS ${_ggml_vk_shader_deps} ${_ggml_vk_genshaders_cmd}
+        DEPENDS ${_ggml_vk_shader_deps}
         COMMENT "Generate vulkan shaders"
     )
 

ggml/src/ggml-vulkan/cmake/host-toolchain.cmake.in

@@ -0,0 +1,15 @@
+set(CMAKE_BUILD_TYPE Release)
+set(CMAKE_C_FLAGS -O2)
+set(CMAKE_CXX_FLAGS -O2)
+set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
+set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY NEVER)
+set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE NEVER)
+set(CMAKE_C_COMPILER @HOST_C_COMPILER@)
+set(CMAKE_CXX_COMPILER @HOST_CXX_COMPILER@)
+set(CMAKE_RUNTIME_OUTPUT_DIRECTORY @CMAKE_RUNTIME_OUTPUT_DIRECTORY@)
+
+if("@CMAKE_C_COMPILER_ID@" STREQUAL "MSVC")
+    foreach(CONFIG IN ITEMS DEBUG RELEASE MINSIZEREL RELWITHDEBINFO)
+        set(CMAKE_RUNTIME_OUTPUT_DIRECTORY_${CONFIG} ${CMAKE_RUNTIME_OUTPUT_DIRECTORY})
+    endforeach()
+endif()

ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt

@@ -1,9 +1,11 @@
 find_package (Threads REQUIRED)
-find_package(Vulkan COMPONENTS glslc REQUIRED)
+find_program(GLSLC_EXECUTABLE glslc)
+if(NOT GLSLC_EXECUTABLE)
+    message(FATAL_ERROR "glslc not found.")
+endif()
 
 set(TARGET vulkan-shaders-gen)
 add_executable(${TARGET} vulkan-shaders-gen.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_compile_features(${TARGET} PRIVATE cxx_std_17)
 target_link_libraries(vulkan-shaders-gen PUBLIC Threads::Threads)
-target_link_libraries(vulkan-shaders-gen PRIVATE Vulkan::Vulkan)

ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp

@@ -30,8 +30,6 @@
     #include <fcntl.h>
 #endif
 
-#include <vulkan/vulkan_core.h>
-
 #define ASYNCIO_CONCURRENCY 64
 
 std::mutex lock;

ggml/src/ggml.c

@@ -3695,7 +3695,7 @@ void ggml_rope_yarn_corr_dims(
 
 // ggml_rope_back
 
-struct ggml_tensor * ggml_rope_back(
+struct ggml_tensor * ggml_rope_ext_back(
         struct ggml_context * ctx,
         struct ggml_tensor  * a,
         struct ggml_tensor  * b,
@@ -3709,29 +3709,32 @@ struct ggml_tensor * ggml_rope_back(
         float                 attn_factor,
         float                 beta_fast,
         float                 beta_slow) {
-    GGML_ASSERT(ggml_is_vector(b));
-    GGML_ASSERT(b->type == GGML_TYPE_I32);
-    GGML_ASSERT(a->ne[2] == b->ne[0]);
-
-    struct ggml_tensor * result = ggml_dup_tensor(ctx, a);
-
-    int32_t params[11] = { /*n_past*/ 0, n_dims, mode, /*n_ctx*/ 0, n_ctx_orig };
-    memcpy(params +  5, &freq_base,    sizeof(float));
-    memcpy(params +  6, &freq_scale,   sizeof(float));
-    memcpy(params +  7, &ext_factor,   sizeof(float));
-    memcpy(params +  8, &attn_factor,  sizeof(float));
-    memcpy(params +  9, &beta_fast,    sizeof(float));
-    memcpy(params + 10, &beta_slow,    sizeof(float));
-    ggml_set_op_params(result, params, sizeof(params));
-
-    result->op     = GGML_OP_ROPE_BACK;
-    result->src[0] = a;
-    result->src[1] = b;
-    result->src[2] = c;
-
+    struct ggml_tensor * result = ggml_rope_ext(
+        ctx, a, b, c, n_dims, mode, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow);
+    result->op = GGML_OP_ROPE_BACK;
     return result;
 }
 
+struct ggml_tensor * ggml_rope_multi_back(
+        struct ggml_context * ctx,
+        struct ggml_tensor  * a,
+        struct ggml_tensor  * b,
+        struct ggml_tensor  * c,
+        int                   n_dims,
+        int                   sections[4],
+        int                   mode,
+        int                   n_ctx_orig,
+        float                 freq_base,
+        float                 freq_scale,
+        float                 ext_factor,
+        float                 attn_factor,
+        float                 beta_fast,
+        float                 beta_slow) {
+    struct ggml_tensor * result = ggml_rope_multi(
+        ctx, a, b, c, n_dims, sections, mode, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow);
+    result->op = GGML_OP_ROPE_BACK;
+    return result;
+}
 // ggml_clamp
 
 struct ggml_tensor * ggml_clamp(
@@ -5594,6 +5597,7 @@ static void ggml_compute_backward(
                 //const int n_ctx      = ((int32_t *) tensor->op_params)[3];
                 const int n_ctx_orig = ((const int32_t *) tensor->op_params)[4];
                 float freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow;
+                int sections[4] = {0, 0, 0, 0};
 
                 memcpy(&freq_base,   (const float *) tensor->op_params +  5, sizeof(float));
                 memcpy(&freq_scale,  (const float *) tensor->op_params +  6, sizeof(float));
@@ -5601,10 +5605,14 @@ static void ggml_compute_backward(
                 memcpy(&attn_factor, (const float *) tensor->op_params +  8, sizeof(float));
                 memcpy(&beta_fast,   (const float *) tensor->op_params +  9, sizeof(float));
                 memcpy(&beta_slow,   (const float *) tensor->op_params + 10, sizeof(float));
+                memcpy(&sections,                    tensor->op_params + 11, sizeof(sections));
 
-                ggml_add_or_set(ctx, cgraph, isrc0,
-                    ggml_rope_back(ctx, grad, src1, src2, n_dims, mode, n_ctx_orig, freq_base,
-                        freq_scale, ext_factor, attn_factor, beta_fast, beta_slow));
+                struct ggml_tensor * rope_back = grad->ne[2] == src1->ne[0] ?
+                    ggml_rope_ext_back(ctx, grad, src1, src2, n_dims,
+                        mode, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow) :
+                    ggml_rope_multi_back(ctx, grad, src1, src2, n_dims, sections,
+                        mode, n_ctx_orig, freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow);
+                ggml_add_or_set(ctx, cgraph, isrc0, rope_back);
             }
             GGML_ASSERT((!src2 || !src2_needs_grads) && "gradients for freq factors not implemented");
         } break;

src/llama.cpp

@@ -4642,7 +4642,7 @@ struct llm_build_context {
                     0);
                 cb(v_states, "v_states", il);
 
-                q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
+                q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend used to not support non-cont. RoPE, investigate removing this
                 q_pe = ggml_rope_ext(
                     ctx0, q_pe, inp_pos, rope_factors,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
@@ -4651,7 +4651,7 @@ struct llm_build_context {
                 cb(q_pe, "q_pe", il);
 
                 // shared RoPE key
-                k_pe = ggml_cont(ctx0, k_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
+                k_pe = ggml_cont(ctx0, k_pe); // TODO: the CUDA backend used to not support non-cont. RoPE, investigate removing this
                 k_pe = ggml_rope_ext(
                     ctx0, k_pe, inp_pos, rope_factors,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
@@ -6496,7 +6496,7 @@ struct llm_build_context {
                     0);
                 cb(v_states, "v_states", il);
 
-                q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
+                q_pe = ggml_cont(ctx0, q_pe); // TODO: the CUDA backend used to not support non-cont. RoPE, investigate removing this
                 q_pe = ggml_rope_ext(
                     ctx0, q_pe, inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
@@ -6505,7 +6505,7 @@ struct llm_build_context {
                 cb(q_pe, "q_pe", il);
 
                 // shared RoPE key
-                k_pe = ggml_cont(ctx0, k_pe); // TODO: the CUDA backend does not support non-contiguous RoPE
+                k_pe = ggml_cont(ctx0, k_pe); // TODO: the CUDA backend used to not support non-cont. RoPE, investigate removing this
                 k_pe = ggml_rope_ext(
                     ctx0, k_pe, inp_pos, nullptr,
                     n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,

tests/test-backend-ops.cpp

@@ -2192,7 +2192,7 @@ struct test_soft_max : public test_case {
 };
 
 
-// GGML_OP_ROPE
+// GGML_OP_ROPE + GGML_OP_ROPE_BACK
 struct test_rope : public test_case {
     const ggml_type type;
     const std::array<int64_t, 4> ne_a;
@@ -2204,29 +2204,36 @@ struct test_rope : public test_case {
     float af; // attn_factor
     bool ff;
     int v; // view (1 : non-contiguous a)
+    bool forward;
 
     std::string vars() override {
+        // forward can be inferred from the op, does not need to be printed
         return VARS_TO_STR10(type, ne_a, n_dims, mode, n_ctx, fs, ef, af, ff, v);
     }
 
     test_rope(ggml_type type = GGML_TYPE_F32,
             std::array<int64_t, 4> ne_a = {10, 5, 3, 1},
-            int n_dims = 10, int mode = 0, int n_ctx = 512, float fs = 1.0f, float ef = 0.0f, float af = 0.0f, bool ff = false, int v = 0)
-        : type(type), ne_a(ne_a), n_dims(n_dims), mode(mode), n_ctx(n_ctx), fs(fs), ef(ef), af(af), ff(ff), v(v) {}
+            int n_dims = 10, int mode = 0, int n_ctx = 512, float fs = 1.0f,
+            float ef = 0.0f, float af = 0.0f, bool ff = false, int v = 0, bool forward = true)
+        : type(type), ne_a(ne_a), n_dims(n_dims), mode(mode), n_ctx(n_ctx), fs(fs), ef(ef), af(af), ff(ff), v(v), forward(forward) {}
 
     ggml_tensor * build_graph(ggml_context * ctx) override {
         ggml_tensor * a;
         if (v & 1) {
             auto ne = ne_a; ne[0] *= 2; ne[1] *= 4; ne[2] *= 3;
             a = ggml_new_tensor(ctx, type, 4, ne.data());
-            ggml_set_param(ctx, a);
+            if (forward) {
+                ggml_set_param(ctx, a);
+            }
             ggml_set_name(a, "a");
 
             a = ggml_view_4d(ctx, a, ne_a[0], ne_a[1], ne_a[2], ne_a[3], a->nb[1], a->nb[2], a->nb[3], 0);
             ggml_set_name(a, "view_of_a");
         } else {
             a = ggml_new_tensor(ctx, type, 4, ne_a.data());
-            ggml_set_param(ctx, a);
+            if (forward) {
+                ggml_set_param(ctx, a);
+            }
             ggml_set_name(a, "a");
         }
 
@@ -2252,14 +2259,26 @@ struct test_rope : public test_case {
             if (is_vision) {
                 GGML_ASSERT(n_dims/4 > 0);
                 int rope_sections[4] = {n_dims/4, n_dims/4, 0, 0}; // Vision-RoPE only use first two dimension for image (x, y) coordinate
-                out = ggml_rope_multi(ctx, a, pos, freq, n_dims/2, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
+                if (forward) {
+                    out = ggml_rope_multi     (ctx, a, pos, freq, n_dims/2, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
+                } else {
+                    out = ggml_rope_multi_back(ctx, a, pos, freq, n_dims/2, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
+                }
             } else {
                 GGML_ASSERT(n_dims/3 > 0);
                 int rope_sections[4] = {n_dims/3, n_dims/3, n_dims/3, 0};
-                out = ggml_rope_multi(ctx, a, pos, freq, n_dims, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
+                if (forward) {
+                    out = ggml_rope_multi     (ctx, a, pos, freq, n_dims, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
+                } else {
+                    out = ggml_rope_multi_back(ctx, a, pos, freq, n_dims, rope_sections, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
+                }
             }
         } else {
-            out = ggml_rope_ext(ctx, a, pos, freq, n_dims, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
+            if (forward) {
+                out = ggml_rope_ext     (ctx, a, pos, freq, n_dims, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
+            } else {
+                out = ggml_rope_ext_back(ctx, a, pos, freq, n_dims, mode, 0, 10000.0f, fs, ef, af, 1.0f, 1.0f);
+            }
         }
         ggml_set_name(out, "out");
 
@@ -3844,7 +3863,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
     test_cases.emplace_back(new test_soft_max(GGML_TYPE_F32, {32, 2, 32, 1}, true,  0.1f, 0.0f));
     test_cases.emplace_back(new test_soft_max(GGML_TYPE_F32, {32, 2, 32, 1}, true,  0.1f, 8.0f));
 
-    {
+    for (bool fw : {true, false}) { // fw == forward
         bool all = true;
 
         for (float v : { 0, 1 }) {
@@ -3853,29 +3872,29 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
                     for (float af : { 1.0f, 1.4245f }) {
                         for (ggml_type type : {GGML_TYPE_F32, GGML_TYPE_F16}) {
                             for (bool ff : {false, true}) { // freq_factors
-                                test_cases.emplace_back(new test_rope(type, {128,  32, 2, 1}, 128, 0, 512, fs, ef, af, ff, v)); // llama 7B
+                                test_cases.emplace_back(new test_rope(type, {128,  32, 2, 1}, 128, 0, 512, fs, ef, af, ff, v, fw)); // llama 7B
 
                                 if (all) {
-                                    test_cases.emplace_back(new test_rope(type, {128,  40, 2, 1}, 128, 0, 512, fs, ef, af, ff, v)); // llama 13B
-                                    test_cases.emplace_back(new test_rope(type, {128,  52, 2, 1}, 128, 0, 512, fs, ef, af, ff, v)); // llama 30B
-                                    test_cases.emplace_back(new test_rope(type, {128,  64, 2, 1}, 128, 0, 512, fs, ef, af, ff, v)); // llama 65B
+                                    test_cases.emplace_back(new test_rope(type, {128,  40, 2, 1}, 128, 0, 512, fs, ef, af, ff, v, fw)); // llama 13B
+                                    test_cases.emplace_back(new test_rope(type, {128,  52, 2, 1}, 128, 0, 512, fs, ef, af, ff, v, fw)); // llama 30B
+                                    test_cases.emplace_back(new test_rope(type, {128,  64, 2, 1}, 128, 0, 512, fs, ef, af, ff, v, fw)); // llama 65B
                                 }
 
                                 if (all) {
-                                    test_cases.emplace_back(new test_rope(type, { 64,   1, 2, 1},  64, 2, 512, fs, ef, af, ff, v)); // neox (falcon 7B)
-                                    test_cases.emplace_back(new test_rope(type, { 64,  71, 2, 1},  64, 2, 512, fs, ef, af, ff, v)); // neox (falcon 7B)
-                                    test_cases.emplace_back(new test_rope(type, { 64,   8, 2, 1},  64, 2, 512, fs, ef, af, ff, v)); // neox (falcon 40B)
-                                    test_cases.emplace_back(new test_rope(type, { 80,  32, 2, 1},  20, 2, 512, fs, ef, af, ff, v)); // neox (stablelm)
-                                    test_cases.emplace_back(new test_rope(type, { 80,  32, 2, 1},  32, 2, 512, fs, ef, af, ff, v)); // neox (phi-2)
+                                    test_cases.emplace_back(new test_rope(type, { 64,   1, 2, 1},  64, 2, 512, fs, ef, af, ff, v, fw)); // neox (falcon 7B)
+                                    test_cases.emplace_back(new test_rope(type, { 64,  71, 2, 1},  64, 2, 512, fs, ef, af, ff, v, fw)); // neox (falcon 7B)
+                                    test_cases.emplace_back(new test_rope(type, { 64,   8, 2, 1},  64, 2, 512, fs, ef, af, ff, v, fw)); // neox (falcon 40B)
+                                    test_cases.emplace_back(new test_rope(type, { 80,  32, 2, 1},  20, 2, 512, fs, ef, af, ff, v, fw)); // neox (stablelm)
+                                    test_cases.emplace_back(new test_rope(type, { 80,  32, 2, 1},  32, 2, 512, fs, ef, af, ff, v, fw)); // neox (phi-2)
                                 }
 
                                 if (all) {
-                                    test_cases.emplace_back(new test_rope(type, {128,  12, 2, 1}, 128, GGML_ROPE_TYPE_MROPE,  512, fs, ef, af, ff, v)); // rope_multi,m-rope (qwen2vl 2B)
-                                    test_cases.emplace_back(new test_rope(type, {128,  28, 2, 1}, 128, GGML_ROPE_TYPE_MROPE,  512, fs, ef, af, ff, v)); // rope_multi,m-rope (qwen2vl 7B)
-                                    test_cases.emplace_back(new test_rope(type, { 80,  16, 2, 1},  80, GGML_ROPE_TYPE_VISION, 512, fs, ef, af, ff, v)); // rope_multi,m-rope (qwen2vl ViT)
+                                    test_cases.emplace_back(new test_rope(type, {128,  12, 2, 1}, 128, GGML_ROPE_TYPE_MROPE,  512, fs, ef, af, ff, v, fw)); // rope_multi,m-rope (qwen2vl 2B)
+                                    test_cases.emplace_back(new test_rope(type, {128,  28, 2, 1}, 128, GGML_ROPE_TYPE_MROPE,  512, fs, ef, af, ff, v, fw)); // rope_multi,m-rope (qwen2vl 7B)
+                                    test_cases.emplace_back(new test_rope(type, { 80,  16, 2, 1},  80, GGML_ROPE_TYPE_VISION, 512, fs, ef, af, ff, v, fw)); // rope_multi,m-rope (qwen2vl ViT)
                                 }
 
-                                test_cases.emplace_back(new test_rope(type, { 64, 128, 2, 1},  64, 2, 512, fs, ef, af, ff, v)); // neox (falcon 40B)
+                                test_cases.emplace_back(new test_rope(type, { 64, 128, 2, 1},  64, 2, 512, fs, ef, af, ff, v, fw)); // neox (falcon 40B)
                             }
                         }