llava : change API to pure C style for Rust FFI bindgen (#6079)

Co-authored-by: Lou Ting <louting.t@alibaba-inc.com>
This commit is contained in:
Ting Lou 2024-03-15 22:31:05 +08:00 committed by GitHub
parent 3020327f6c
commit 4e9a7f7f7f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 24 additions and 24 deletions

View File

@ -1235,16 +1235,16 @@ struct clip_image_f32 * clip_image_f32_init() {
void clip_image_u8_free(struct clip_image_u8 * img) { delete img; } void clip_image_u8_free(struct clip_image_u8 * img) { delete img; }
void clip_image_f32_free(struct clip_image_f32 * img) { delete img; } void clip_image_f32_free(struct clip_image_f32 * img) { delete img; }
void clip_image_u8_batch_free(struct clip_image_u8_batch & batch) { void clip_image_u8_batch_free(struct clip_image_u8_batch * batch) {
if (batch.size > 0) { if (batch->size > 0) {
delete[] batch.data; delete[] batch->data;
batch.size = 0; batch->size = 0;
} }
} }
void clip_image_f32_batch_free(struct clip_image_f32_batch & batch) { void clip_image_f32_batch_free(struct clip_image_f32_batch * batch) {
if (batch.size > 0) { if (batch->size > 0) {
delete[] batch.data; delete[] batch->data;
batch.size = 0; batch->size = 0;
} }
} }
@ -1497,7 +1497,7 @@ static std::vector<clip_image_u8*> divide_to_patches_u8(const clip_image_u8 & im
// returns the normalized float tensor for llava-1.5, for spatial_unpad with anyres processing for llava-1.6 it returns the normalized image patch tensors as a vector // returns the normalized float tensor for llava-1.5, for spatial_unpad with anyres processing for llava-1.6 it returns the normalized image patch tensors as a vector
// res_imgs memory is being allocated here, previous allocations will be freed if found // res_imgs memory is being allocated here, previous allocations will be freed if found
bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, clip_image_f32_batch & res_imgs) { bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, clip_image_f32_batch * res_imgs) {
bool pad_to_square = true; bool pad_to_square = true;
if (!ctx->has_vision_encoder) { if (!ctx->has_vision_encoder) {
printf("This gguf file seems to have no vision encoder\n"); printf("This gguf file seems to have no vision encoder\n");
@ -1509,11 +1509,11 @@ bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, cli
pad_to_square = false; pad_to_square = false;
} }
// free the previous res_imgs if any set // free the previous res_imgs if any set
if (res_imgs.size > 0) { if (res_imgs->size > 0) {
clip_image_f32_batch_free(res_imgs); clip_image_f32_batch_free(res_imgs);
} }
res_imgs.data = nullptr; res_imgs->data = nullptr;
res_imgs.size = 0; res_imgs->size = 0;
// the logic below is to pad the shorter side to the longer side with a background color: rgb(122, 116, 104) // the logic below is to pad the shorter side to the longer side with a background color: rgb(122, 116, 104)
// see https://github.com/haotian-liu/LLaVA/blob/e854a2bf85118c504f6f16bf5c3c7c92f8fa8c6b/llava/conversation.py#L113-L156 // see https://github.com/haotian-liu/LLaVA/blob/e854a2bf85118c504f6f16bf5c3c7c92f8fa8c6b/llava/conversation.py#L113-L156
@ -1568,11 +1568,11 @@ bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, cli
bicubic_resize(*img, *image_original_resize, params.image_size, params.image_size); // in python this is "shortest_edge", but all CLIP are square bicubic_resize(*img, *image_original_resize, params.image_size, params.image_size); // in python this is "shortest_edge", but all CLIP are square
patches.insert(patches.begin(), image_original_resize); patches.insert(patches.begin(), image_original_resize);
// clip_image_f32_batch_init(patches.size()); // clip_image_f32_batch_init(patches.size());
res_imgs.size = patches.size(); res_imgs->size = patches.size();
res_imgs.data = new clip_image_f32[res_imgs.size]; res_imgs->data = new clip_image_f32[res_imgs->size];
int num=0; int num=0;
for (auto& patch : patches) { for (auto& patch : patches) {
normalize_image_u8_to_f32(patch, &res_imgs.data[num], ctx->image_mean, ctx->image_std); normalize_image_u8_to_f32(patch, &res_imgs->data[num], ctx->image_mean, ctx->image_std);
num++; num++;
} }
@ -1660,9 +1660,9 @@ bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, cli
// } // }
// res_imgs.push_back(res); // res_imgs.push_back(res);
res_imgs.size = 1; res_imgs->size = 1;
res_imgs.data = new clip_image_f32[res_imgs.size]; res_imgs->data = new clip_image_f32[res_imgs->size];
res_imgs.data[0] = *res; res_imgs->data[0] = *res;
clip_image_f32_free(res); clip_image_f32_free(res);
return true; return true;

View File

@ -60,8 +60,8 @@ CLIP_API struct clip_image_f32 * clip_image_f32_init();
CLIP_API void clip_image_u8_free (struct clip_image_u8 * img); CLIP_API void clip_image_u8_free (struct clip_image_u8 * img);
CLIP_API void clip_image_f32_free(struct clip_image_f32 * img); CLIP_API void clip_image_f32_free(struct clip_image_f32 * img);
CLIP_API void clip_image_u8_batch_free (struct clip_image_u8_batch & batch); CLIP_API void clip_image_u8_batch_free (struct clip_image_u8_batch * batch);
CLIP_API void clip_image_f32_batch_free(struct clip_image_f32_batch & batch); CLIP_API void clip_image_f32_batch_free(struct clip_image_f32_batch * batch);
CLIP_API bool clip_image_load_from_file(const char * fname, struct clip_image_u8 * img); CLIP_API bool clip_image_load_from_file(const char * fname, struct clip_image_u8 * img);
@ -69,7 +69,7 @@ CLIP_API bool clip_image_load_from_file(const char * fname, struct clip_image_u8
CLIP_API bool clip_image_load_from_bytes(const unsigned char * bytes, size_t bytes_length, struct clip_image_u8 * img); CLIP_API bool clip_image_load_from_bytes(const unsigned char * bytes, size_t bytes_length, struct clip_image_u8 * img);
/** preprocess img and store the result in res_imgs, pad_to_square may be overriden to false depending on model configuration */ /** preprocess img and store the result in res_imgs, pad_to_square may be overriden to false depending on model configuration */
CLIP_API bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, clip_image_f32_batch & res_imgs ); CLIP_API bool clip_image_preprocess(struct clip_ctx * ctx, const struct clip_image_u8 * img, struct clip_image_f32_batch * res_imgs );
CLIP_API struct ggml_tensor * clip_get_newline_tensor(const struct clip_ctx * ctx); CLIP_API struct ggml_tensor * clip_get_newline_tensor(const struct clip_ctx * ctx);

View File

@ -223,7 +223,7 @@ static bool encode_image_with_clip(clip_ctx * ctx_clip, int n_threads, const cli
clip_image_f32_batch img_res_v; clip_image_f32_batch img_res_v;
img_res_v.size = 0; img_res_v.size = 0;
img_res_v.data = nullptr; img_res_v.data = nullptr;
if (!clip_image_preprocess(ctx_clip, img, img_res_v)) { if (!clip_image_preprocess(ctx_clip, img, &img_res_v)) {
fprintf(stderr, "%s: unable to preprocess image\n", __func__); fprintf(stderr, "%s: unable to preprocess image\n", __func__);
delete[] img_res_v.data; delete[] img_res_v.data;
return false; return false;

View File

@ -29,9 +29,9 @@ struct llava_image_embed {
}; };
/** sanity check for clip <-> llava embed size match */ /** sanity check for clip <-> llava embed size match */
LLAVA_API bool llava_validate_embed_size(const llama_context * ctx_llama, const clip_ctx * ctx_clip); LLAVA_API bool llava_validate_embed_size(const struct llama_context * ctx_llama, const struct clip_ctx * ctx_clip);
LLAVA_API bool llava_image_embed_make_with_clip_img(clip_ctx * ctx_clip, int n_threads, const clip_image_u8 * img, float ** image_embd_out, int * n_img_pos_out); LLAVA_API bool llava_image_embed_make_with_clip_img(struct clip_ctx * ctx_clip, int n_threads, const struct clip_image_u8 * img, float ** image_embd_out, int * n_img_pos_out);
/** build an image embed from image file bytes */ /** build an image embed from image file bytes */
LLAVA_API struct llava_image_embed * llava_image_embed_make_with_bytes(struct clip_ctx * ctx_clip, int n_threads, const unsigned char * image_bytes, int image_bytes_length); LLAVA_API struct llava_image_embed * llava_image_embed_make_with_bytes(struct clip_ctx * ctx_clip, int n_threads, const unsigned char * image_bytes, int image_bytes_length);