diff --git a/Makefile b/Makefile index 0a2070b539df8..083a639c52ac2 100644 --- a/Makefile +++ b/Makefile @@ -691,7 +691,7 @@ save-load-state: examples/save-load-state/save-load-state.cpp ggml.o llama.o $(C $(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<) $(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS) -server: examples/server/server.cpp examples/server/oai.hpp examples/server/utils.hpp examples/server/httplib.h examples/server/json.hpp examples/server/index.html.hpp examples/server/index.js.hpp examples/server/completion.js.hpp examples/llava/clip.cpp examples/llava/clip.h common/stb_image.h ggml.o llama.o $(COMMON_DEPS) grammar-parser.o $(OBJS) +server: examples/server/server.cpp examples/server/oai.hpp examples/server/utils.hpp examples/server/httplib.h examples/server/json.hpp examples/server/index.html.hpp examples/server/index.js.hpp examples/server/completion.js.hpp examples/llava/clip.cpp examples/llava/clip.h examples/llava/llava.h examples/llava/llava.cpp common/stb_image.h ggml.o llama.o $(COMMON_DEPS) grammar-parser.o $(OBJS) $(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<) $(CXX) $(CXXFLAGS) -c examples/llava/clip.cpp -o $(call GET_OBJ_FILE, examples/llava/clip.cpp) -Wno-cast-qual $(CXX) $(CXXFLAGS) -Iexamples/server $(filter-out %.h %.hpp $< examples/llava/clip.cpp,$^) $(call GET_OBJ_FILE, $<) $(call GET_OBJ_FILE, examples/llava/clip.cpp) -o $@ $(LDFLAGS) $(LWINSOCK2) diff --git a/build.zig b/build.zig index 699738f3dd509..c0af454dc9e92 100644 --- a/build.zig +++ b/build.zig @@ -123,6 +123,7 @@ pub fn build(b: *std.build.Builder) !void { const grammar_parser = make.obj("grammar-parser", "common/grammar-parser.cpp"); const train = make.obj("train", "common/train.cpp"); const clip = make.obj("clip", "examples/llava/clip.cpp"); + const llava = make.obj("llava", "examples/llava/llava.cpp"); _ = make.exe("main", "examples/main/main.cpp", &.{ ggml, ggml_alloc, ggml_backend, ggml_quants, llama, common, buildinfo, sampling, console, grammar_parser }); _ = make.exe("quantize", "examples/quantize/quantize.cpp", &.{ ggml, ggml_alloc, ggml_backend, ggml_quants, llama, common, buildinfo }); @@ -131,7 +132,7 @@ pub fn build(b: *std.build.Builder) !void { _ = make.exe("finetune", "examples/finetune/finetune.cpp", &.{ ggml, ggml_alloc, ggml_backend, ggml_quants, llama, common, buildinfo, train }); _ = make.exe("train-text-from-scratch", "examples/train-text-from-scratch/train-text-from-scratch.cpp", &.{ ggml, ggml_alloc, ggml_backend, ggml_quants, llama, common, buildinfo, train }); - const server = make.exe("server", "examples/server/server.cpp", &.{ ggml, ggml_alloc, ggml_backend, ggml_quants, llama, common, buildinfo, sampling, grammar_parser, clip }); + const server = make.exe("server", "examples/server/server.cpp", &.{ ggml, ggml_alloc, ggml_backend, ggml_quants, llama, common, buildinfo, sampling, grammar_parser, clip, llava }); if (server.target.isWindows()) { server.linkSystemLibrary("ws2_32"); } diff --git a/examples/llava/llava.cpp b/examples/llava/llava.cpp index 4cb65a07b6740..1a1cf7c78bf34 100644 --- a/examples/llava/llava.cpp +++ b/examples/llava/llava.cpp @@ -311,7 +311,7 @@ bool llava_validate_embed_size(const llama_context * ctx_llama, const clip_ctx * return true; } -static 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) { +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) { float * image_embd = (float *)malloc(clip_embd_nbytes(ctx_clip)*6); // TODO: base on gridsize/llava model if (!image_embd) { fprintf(stderr, "Unable to allocate memory for image embeddings\n"); diff --git a/examples/llava/llava.h b/examples/llava/llava.h index 9e9466a5d1726..2d40f3f1d5f84 100644 --- a/examples/llava/llava.h +++ b/examples/llava/llava.h @@ -31,6 +31,8 @@ struct llava_image_embed { /** 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_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); + /** 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); /** build an image embed from a path to an image filename */ diff --git a/examples/server/server.cpp b/examples/server/server.cpp index a0b46970b83a9..c4cd325f5231f 100644 --- a/examples/server/server.cpp +++ b/examples/server/server.cpp @@ -5,6 +5,7 @@ #include "oai.hpp" #include "../llava/clip.h" +#include "../llava/llava.h" #include "stb_image.h" @@ -983,43 +984,12 @@ struct llama_server_context { continue; } - clip_image_f32_batch img_res_v; - img_res_v.size = 0; - img_res_v.data = nullptr; - if (!clip_image_preprocess(clp_ctx, img.img_data, img_res_v)) - { - LOG_TEE("Error processing the given image"); - clip_free(clp_ctx); - clip_image_f32_batch_free(img_res_v); - return false; - } - if (img_res_v.size == 0) - { - LOG_TEE("Error processing the given image"); - return false; - } - - // note: assumes only one image was returned by clip_image_preprocess - clip_image_f32 * img_res = img_res_v.data; - img.image_tokens = clip_n_patches(clp_ctx); - img.image_embedding = (float *)malloc(clip_embd_nbytes(clp_ctx)); - if (!img.image_embedding) - { - LOG_TEE("Unable to allocate memory for image embeddings\n"); - clip_image_f32_batch_free(img_res_v); - clip_free(clp_ctx); - return false; - } - LOG_TEE("slot %i - encoding image [id: %i]\n", slot.id, img.id); - if (!clip_image_encode(clp_ctx, params.n_threads, img_res, img.image_embedding)) - { - LOG_TEE("Unable to encode image\n"); - clip_image_f32_batch_free(img_res_v); + if (!llava_image_embed_make_with_clip_img(clp_ctx, params.n_threads, img.img_data, &img.image_embedding, &img.image_tokens)) { + LOG_TEE("Error processing the given image"); return false; } - clip_image_f32_batch_free(img_res_v); img.request_encode_image = false; }