ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

common : refactor downloading system, handle mmproj with -hf option (#12694)

Browse source 
* (wip) refactor downloading system [no ci]

* fix all examples

* fix mmproj with -hf

* gemma3: update readme

* only handle mmproj in llava example

* fix multi-shard download

* windows: fix problem with std::min and std::max

* fix 2
This commit is contained in:
Xuan-Son Nguyen 2025-04-01 23:44:05 +02:00 • committed by GitHub
parent f423981ac8
commit 267c1399f1
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
19 changed files with 673 additions and 635 deletions
Download patch file Download diff file Expand all files Collapse all files

20
examples/llava/README-gemma3.md
View file

@ -4,6 +4,26 @@
>
> This is very experimental, only used for demo purpose.
## Quick started
You can use pre-quantized model from [ggml-org](https://huggingface.co/ggml-org)'s Hugging Face account
```bash
# build
cmake -B build
cmake --build build --target llama-gemma3-cli
# alternatively, install from brew (MacOS)
brew install llama.cpp
# run it
llama-gemma3-cli -hf ggml-org/gemma-3-4b-it-GGUF
llama-gemma3-cli -hf ggml-org/gemma-3-12b-it-GGUF
llama-gemma3-cli -hf ggml-org/gemma-3-27b-it-GGUF
# note: 1B model does not support vision
```
## How to get mmproj.gguf?
```bash

6
examples/llava/gemma3-cli.cpp
View file

@ -78,7 +78,7 @@ struct gemma3_context {
}
void init_clip_model(common_params & params) {
const char * clip_path = params.mmproj.c_str();
const char * clip_path = params.mmproj.path.c_str();
ctx_clip = clip_model_load(clip_path, params.verbosity > 1);
}
@ -232,13 +232,13 @@ int main(int argc, char ** argv) {
common_init();
if (params.mmproj.empty()) {
if (params.mmproj.path.empty()) {
show_additional_info(argc, argv);
return 1;
}
gemma3_context ctx(params);
printf("%s: %s\n", __func__, params.model.c_str());
printf("%s: %s\n", __func__, params.model.path.c_str());
bool is_single_turn = !params.prompt.empty() && !params.image.empty();

6
examples/llava/llava-cli.cpp
View file

@ -225,7 +225,7 @@ static struct llama_model * llava_init(common_params * params) {
llama_model_params model_params = common_model_params_to_llama(*params);
llama_model * model = llama_model_load_from_file(params->model.c_str(), model_params);
llama_model * model = llama_model_load_from_file(params->model.path.c_str(), model_params);
if (model == NULL) {
LOG_ERR("%s: unable to load model\n" , __func__);
return NULL;
@ -234,7 +234,7 @@ static struct llama_model * llava_init(common_params * params) {
}
static struct llava_context * llava_init_context(common_params * params, llama_model * model) {
const char * clip_path = params->mmproj.c_str();
const char * clip_path = params->mmproj.path.c_str();
auto prompt = params->prompt;
if (prompt.empty()) {
@ -283,7 +283,7 @@ int main(int argc, char ** argv) {
common_init();
if (params.mmproj.empty() || (params.image.empty() && !prompt_contains_image(params.prompt))) {
if (params.mmproj.path.empty() || (params.image.empty() && !prompt_contains_image(params.prompt))) {
print_usage(argc, argv);
return 1;
}

6
examples/llava/minicpmv-cli.cpp
View file

@ -31,7 +31,7 @@ static struct llama_model * llava_init(common_params * params) {
llama_model_params model_params = common_model_params_to_llama(*params);
llama_model * model = llama_model_load_from_file(params->model.c_str(), model_params);
llama_model * model = llama_model_load_from_file(params->model.path.c_str(), model_params);
if (model == NULL) {
LOG_ERR("%s: unable to load model\n" , __func__);
return NULL;
@ -80,7 +80,7 @@ static void llava_free(struct llava_context * ctx_llava) {
}
static struct clip_ctx * clip_init_context(common_params * params) {
const char * clip_path = params->mmproj.c_str();
const char * clip_path = params->mmproj.path.c_str();
auto prompt = params->prompt;
if (prompt.empty()) {
@ -290,7 +290,7 @@ int main(int argc, char ** argv) {
common_init();
if (params.mmproj.empty() || (params.image.empty())) {
if (params.mmproj.path.empty() || (params.image.empty())) {
show_additional_info(argc, argv);
return 1;
}

6
examples/llava/qwen2vl-cli.cpp
View file

@ -314,7 +314,7 @@ static struct llama_model * llava_init(common_params * params) {
llama_model_params model_params = common_model_params_to_llama(*params);
llama_model * model = llama_model_load_from_file(params->model.c_str(), model_params);
llama_model * model = llama_model_load_from_file(params->model.path.c_str(), model_params);
if (model == NULL) {
LOG_ERR("%s: unable to load model\n" , __func__);
return NULL;
@ -323,7 +323,7 @@ static struct llama_model * llava_init(common_params * params) {
}
static struct llava_context * llava_init_context(common_params * params, llama_model * model) {
const char * clip_path = params->mmproj.c_str();
const char * clip_path = params->mmproj.path.c_str();
auto prompt = params->prompt;
if (prompt.empty()) {
@ -524,7 +524,7 @@ int main(int argc, char ** argv) {
common_init();
if (params.mmproj.empty() || (params.image.empty() && !prompt_contains_image(params.prompt))) {
if (params.mmproj.path.empty() || (params.image.empty() && !prompt_contains_image(params.prompt))) {
print_usage(argc, argv);
return 1;
}