root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-28 12:24:35 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

fix llava implementation

Browse Source
This commit is contained in:
FSSRepo 2023-10-16 16:31:17 -04:00
parent d7eca255d7
commit 4d1804330e
3 changed files with 1535 additions and 1527 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3042
examples/server/index.html.hpp
View File

File diff suppressed because it is too large Load Diff

2
examples/server/public/index.html
View File

@ -446,7 +446,7 @@
).join("\n"), ).join("\n"),
}); });
if(selected_image) { if(selected_image) {
prompt = `A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.\nUSER: [img-10]${msg}\nASSISTANT:`; prompt = `A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.\nUSER:[img-10]${msg}\nASSISTANT:`;
} }
await runLlama(prompt, { await runLlama(prompt, {
...params.value, ...params.value,

18
examples/server/server.cpp
View File

@ -285,7 +285,7 @@ struct llama_client_slot
int32_t n_past = 0; int32_t n_past = 0;
int32_t n_decoded = 0; int32_t n_decoded = 0;
int32_t i_batch = -1; int32_t i_batch = -1;
int32_t num_prompt_tokens = 0; size_t num_prompt_tokens = 0;
int32_t num_prompt_tokens_processed = 0; int32_t num_prompt_tokens_processed = 0;
int32_t n_remaining = -1; int32_t n_remaining = -1;
@ -347,6 +347,15 @@ struct llama_client_slot
ctx_sampling.grammar = NULL; ctx_sampling.grammar = NULL;
} }
#ifdef SERVER_MULTIMODAL_SUPPORT
for(slot_image img : images) {
free(img.image_embedding);
delete[] img.img_data.data;
img.prefix_prompt = "";
}
images.clear();
#endif
// llama_set_rng_seed(ctx, params.seed); in batched the seed matter??????? // llama_set_rng_seed(ctx, params.seed); in batched the seed matter???????
} }
@ -893,7 +902,7 @@ struct llama_server_context
const auto json_prompt = (image_idx >= slot.images.size()) ? const auto json_prompt = (image_idx >= slot.images.size()) ?
slot.params.input_suffix : // no more images, then process suffix prompt slot.params.input_suffix : // no more images, then process suffix prompt
(json)(slot.images[image_idx].prefix_prompt); (json)(slot.images[image_idx].prefix_prompt);
std::vector<llama_token> append_tokens = tokenize(json_prompt, true); // has next image std::vector<llama_token> append_tokens = tokenize(json_prompt, false); // has next image
for (int i = 0; i < append_tokens.size(); ++i) { for (int i = 0; i < append_tokens.size(); ++i) {
batch.token [batch.n_tokens] = append_tokens[i]; batch.token [batch.n_tokens] = append_tokens[i];
batch.pos [batch.n_tokens] = slot.n_past; batch.pos [batch.n_tokens] = slot.n_past;
@ -1810,7 +1819,6 @@ static void parse_options_completion(const json &body, llama_client_slot* slot,
const auto &images_data = body.find("image_data"); const auto &images_data = body.find("image_data");
if (images_data != body.end() && images_data->is_array()) if (images_data != body.end() && images_data->is_array())
{ {
slot->images.clear();
for (const auto &img : *images_data) for (const auto &img : *images_data)
{ {
slot_image img_sl; slot_image img_sl;
@ -1821,10 +1829,10 @@ static void parse_options_completion(const json &body, llama_client_slot* slot,
data_b64.clear(); data_b64.clear();
auto data = stbi_load_from_memory(image_buffer.data(), image_buffer.size(), &width, &height, &channels, 3); auto data = stbi_load_from_memory(image_buffer.data(), image_buffer.size(), &width, &height, &channels, 3);
if(!data) { if(!data) {
LOG_TEE("slot %i - failed to load image\n", slot->id); LOG_TEE("slot %i - failed to load image id= %i\n", slot->id, img_sl.id);
return; return;
} }
LOG_TEE("slot %i - RGB image %i loaded (%i x %i)\n", slot->id, img_sl.id, width, height); LOG_TEE("slot %i - image id= %i loaded (%i x %i)\n", slot->id, img_sl.id, width, height);
img_sl.img_data.nx = width; img_sl.img_data.nx = width;
img_sl.img_data.ny = height; img_sl.img_data.ny = height;
img_sl.img_data.size = width * height * 3; img_sl.img_data.size = width * height * 3;