kanshan/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-07-11 05:30:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

MTP: clean-up (#9)

Browse Source 
* MTP: clean-up

* review: use llama_context_type instead of llama_graph_type

* review: remove llama_model_has_mtp

* review: fix convert issues

* convert: fix pycheck

* review: formatting

* use `mtp-` for identifying mtp models

* convert: fix mtp conversion
This commit is contained in:
Aman Gupta
2026-05-13 11:12:20 +08:00
parent 5c58cc5bdd
commit 3c3aebaaa0
20 changed files with 687 additions and 646 deletions
Download Patch File Download Diff File Expand all files Collapse all files
common/arg.cpp
+24 -3
View File
@@ -335,11 +335,15 @@ static bool common_params_handle_remote_preset(common_params & params, llama_exa
struct handle_model_result {
bool found_mmproj = false;
common_params_model mmproj;
bool found_mtp = false;
common_params_model mtp;
};
static handle_model_result common_params_handle_model(struct common_params_model & model,
const std::string & bearer_token,
bool offline) {
bool offline,
bool search_mtp = false) {
handle_model_result result;
if (!model.docker_repo.empty()) {
@@ -354,7 +358,7 @@ static handle_model_result common_params_handle_model(struct common_params_model
common_download_opts opts;
opts.bearer_token = bearer_token;
opts.offline = offline;
auto download_result = common_download_model(model, opts, true);
auto download_result = common_download_model(model, opts, true, search_mtp);
if (download_result.model_path.empty()) {
LOG_ERR("error: failed to download model from Hugging Face\n");
@@ -368,6 +372,11 @@ static handle_model_result common_params_handle_model(struct common_params_model
result.found_mmproj = true;
result.mmproj.path = download_result.mmproj_path;
}
if (!download_result.mtp_path.empty()) {
result.found_mtp = true;
result.mtp.path = download_result.mtp_path;
}
} else if (!model.url.empty()) {
if (model.path.empty()) {
auto f = string_split<std::string>(model.url, '#').front();
@@ -436,7 +445,11 @@ static bool parse_bool_value(const std::string & value) {
//
void common_params_handle_models(common_params & params, llama_example curr_ex) {
auto res = common_params_handle_model(params.model, params.hf_token, params.offline);
const bool spec_type_mtp = std::find(params.speculative.types.begin(),
params.speculative.types.end(),
COMMON_SPECULATIVE_TYPE_MTP) != params.speculative.types.end();
auto res = common_params_handle_model(params.model, params.hf_token, params.offline, spec_type_mtp);
if (params.no_mmproj) {
params.mmproj = {};
} else if (res.found_mmproj && params.mmproj.path.empty() && params.mmproj.url.empty()) {
@@ -450,6 +463,14 @@ void common_params_handle_models(common_params & params, llama_example curr_ex)
break;
}
}
// when --spec-type mtp is set and no draft model was provided explicitly,
// fall back to the MTP head discovered alongside the -hf model
if (spec_type_mtp && res.found_mtp &&
params.speculative.draft.mparams.path.empty() &&
params.speculative.draft.mparams.hf_repo.empty() &&
params.speculative.draft.mparams.url.empty()) {
params.speculative.draft.mparams.path = res.mtp.path;
}
common_params_handle_model(params.speculative.draft.mparams, params.hf_token, params.offline);
common_params_handle_model(params.vocoder.model, params.hf_token, params.offline);
}