vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix the editor config checks

Browse source
This commit is contained in:
Chenxiaotao03 2024-01-22 19:51:14 +08:00
parent c37859bf21
commit 0e57eb875e
3 changed files with 12 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

2
examples/llava/MobileVLM-README.md
View file

@ -128,4 +128,4 @@ The `n_patch` of output in `ldp` is 1/4 of the input. In order to implement quic
## contributor ## contributor
```sh ```sh
zhangjidong05, yangyang260, huyiming03, chenxiaotao03 zhangjidong05, yangyang260, huyiming03, chenxiaotao03
``` ```

4
examples/llava/android/adb_run.sh
View file

@ -29,7 +29,7 @@ function android_run() {
# copy program into device # copy program into device
adb push ${program_dir}/${binName} ${deviceDir}/${binName} adb push ${program_dir}/${binName} ${deviceDir}/${binName}
adb shell "chmod 0777 ${deviceDir}/${binName}" adb shell "chmod 0777 ${deviceDir}/${binName}"
# run # run
adb shell "echo cd ${deviceDir} ${deviceDir}/${binName} \ adb shell "echo cd ${deviceDir} ${deviceDir}/${binName} \
-m ${deviceDir}/${llama_name} \ -m ${deviceDir}/${llama_name} \
@ -50,4 +50,4 @@ function android_run() {
android_run android_run
echo "android_run is Done!" echo "android_run is Done!"

18
examples/llava/clip.cpp
View file

@ -217,8 +217,8 @@ static std::string gguf_kv_to_str(const struct gguf_context * ctx_gguf, int i) {
static void print_tensor_info(const ggml_tensor* tensor, const char* prefix = "") { static void print_tensor_info(const ggml_tensor* tensor, const char* prefix = "") {
size_t tensor_size = ggml_nbytes(tensor); size_t tensor_size = ggml_nbytes(tensor);
printf("%s: n_dims = %d, name = %s, tensor_size=%zu, shape:[%d, %d, %d, %d], type: %d\n", printf("%s: n_dims = %d, name = %s, tensor_size=%zu, shape:[%d, %d, %d, %d], type: %d\n",
prefix, ggml_n_dims(tensor), tensor->name, tensor_size, prefix, ggml_n_dims(tensor), tensor->name, tensor_size,
tensor->ne[0], tensor->ne[1], tensor->ne[2], tensor->ne[3], tensor->type); tensor->ne[0], tensor->ne[1], tensor->ne[2], tensor->ne[3], tensor->type);
} }
@ -593,7 +593,7 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
mlp_3 = ggml_reshape_4d(ctx0, mlp_3, n_patch, n_patch, mlp_3->ne[1], mlp_3->ne[2]); mlp_3 = ggml_reshape_4d(ctx0, mlp_3, n_patch, n_patch, mlp_3->ne[1], mlp_3->ne[2]);
// stride = 1, padding = 1, bias is nullptr // stride = 1, padding = 1, bias is nullptr
block_1 = ggml_conv_depthwise_2d(ctx0, model.mm_model_block_1_block_0_0_w, mlp_3, nullptr, 1, 1, 1, 1, 1, 1); block_1 = ggml_conv_depthwise_2d(ctx0, model.mm_model_block_1_block_0_0_w, mlp_3, nullptr, 1, 1, 1, 1, 1, 1);
// layer norm // layer norm
// // block_1 shape = [1, 2048, 24, 24], ne = [24, 24, 2048, 1] // // block_1 shape = [1, 2048, 24, 24], ne = [24, 24, 2048, 1]
block_1 = ggml_cont(ctx0, ggml_permute(ctx0, block_1, 1, 2, 0, 3)); block_1 = ggml_cont(ctx0, ggml_permute(ctx0, block_1, 1, 2, 0, 3));
@ -601,11 +601,11 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
block_1 = ggml_norm(ctx0, block_1, eps); block_1 = ggml_norm(ctx0, block_1, eps);
block_1 = ggml_add(ctx0, ggml_mul(ctx0, block_1, model.mm_model_block_1_block_0_1_w), model.mm_model_block_1_block_0_1_b); block_1 = ggml_add(ctx0, ggml_mul(ctx0, block_1, model.mm_model_block_1_block_0_1_w), model.mm_model_block_1_block_0_1_b);
block_1 = ggml_cont(ctx0, ggml_permute(ctx0, block_1, 2, 0, 1, 3)); block_1 = ggml_cont(ctx0, ggml_permute(ctx0, block_1, 2, 0, 1, 3));
// block_1 shape = [1, 2048, 24, 24], ne = [24, 24, 2048, 1] // block_1 shape = [1, 2048, 24, 24], ne = [24, 24, 2048, 1]
// hardswish // hardswish
struct ggml_tensor * block_1_hw = ggml_hardswish(ctx0, block_1); struct ggml_tensor * block_1_hw = ggml_hardswish(ctx0, block_1);
block_1 = ggml_pool_2d(ctx0, block_1_hw, GGML_OP_POOL_AVG, block_1_hw->ne[0], block_1_hw->ne[1], block_1_hw->ne[0], block_1_hw->ne[1], 0, 0); block_1 = ggml_pool_2d(ctx0, block_1_hw, GGML_OP_POOL_AVG, block_1_hw->ne[0], block_1_hw->ne[1], block_1_hw->ne[0], block_1_hw->ne[1], 0, 0);
// block_1 shape = [1, 2048, 1, 1], ne = [1, 1, 2048, 1] // block_1 shape = [1, 2048, 1, 1], ne = [1, 1, 2048, 1]
// pointwise conv // pointwise conv
@ -641,7 +641,7 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
{ {
// stride = 2 // stride = 2
block_1 = ggml_conv_depthwise_2d(ctx0, model.mm_model_block_2_block_0_0_w, block_1, nullptr, 2, 2, 1, 1, 1, 1); block_1 = ggml_conv_depthwise_2d(ctx0, model.mm_model_block_2_block_0_0_w, block_1, nullptr, 2, 2, 1, 1, 1, 1);
// block_1 shape = [1, 2048, 12, 12], ne = [12, 12, 2048, 1] // block_1 shape = [1, 2048, 12, 12], ne = [12, 12, 2048, 1]
// layer norm // layer norm
block_1 = ggml_cont(ctx0, ggml_permute(ctx0, block_1, 1, 2, 0, 3)); block_1 = ggml_cont(ctx0, ggml_permute(ctx0, block_1, 1, 2, 0, 3));
@ -679,10 +679,10 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
// block_1 shape = [1, 12, 12, 2048], ne = [2048, 12, 12, 1] // block_1 shape = [1, 12, 12, 2048], ne = [2048, 12, 12, 1]
block_1 = ggml_norm(ctx0, block_1, eps); block_1 = ggml_norm(ctx0, block_1, eps);
block_1 = ggml_add(ctx0, ggml_mul(ctx0, block_1, model.mm_model_block_2_block_2_1_w), model.mm_model_block_2_block_2_1_b); block_1 = ggml_add(ctx0, ggml_mul(ctx0, block_1, model.mm_model_block_2_block_2_1_w), model.mm_model_block_2_block_2_1_b);
block_1 = ggml_reshape_3d(ctx0, block_1, block_1->ne[0], block_1->ne[1] * block_1->ne[2], block_1->ne[3]); block_1 = ggml_reshape_3d(ctx0, block_1, block_1->ne[0], block_1->ne[1] * block_1->ne[2], block_1->ne[3]);
// block_1 shape = [1, 144, 2048], ne = [2048, 144, 1] // block_1 shape = [1, 144, 2048], ne = [2048, 144, 1]
} }
embeddings = block_1; embeddings = block_1;
} }
else { else {
@ -996,7 +996,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
vision_model.mm_model_block_2_block_1_fc2_b = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_BLOCK, 2, 1, "fc2.bias")); vision_model.mm_model_block_2_block_1_fc2_b = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_BLOCK, 2, 1, "fc2.bias"));
vision_model.mm_model_block_2_block_2_0_w = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_BLOCK, 2, 2, "0.weight")); vision_model.mm_model_block_2_block_2_0_w = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_BLOCK, 2, 2, "0.weight"));
vision_model.mm_model_block_2_block_2_1_w = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_BLOCK, 2, 2, "1.weight")); vision_model.mm_model_block_2_block_2_1_w = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_BLOCK, 2, 2, "1.weight"));
vision_model.mm_model_block_2_block_2_1_b = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_BLOCK, 2, 2, "1.bias")); vision_model.mm_model_block_2_block_2_1_b = get_tensor(new_clip->ctx_data, format(TN_MVLM_PROJ_BLOCK, 2, 2, "1.bias"));
} }
else { else {
std::string proj_type = PROJECTOR_TYPE_NAMES[new_clip->proj_type]; std::string proj_type = PROJECTOR_TYPE_NAMES[new_clip->proj_type];