mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 32 Packages Projects Releases Wiki Activity

model: move load_hparams and load_tensors to per-model definition (#22004)

Browse Source 
* git-friendly migration

* add build_graph

* nits

* exclude old code from build

* wip

* add llm_arch_model_i

* prepare downstream functions

* nits

* nits

* wip

* wip

* add back create_tensor_qkv

* fix files missing include

* enforce one llm_build per arch

* cmake: use glob

* missing model params

* nits

* wip

* wip (2)

* wip (3)

* test-llama-archs is happy

* improve switch case

* move more stuff into llm_arch_model_i

* fix downstream code

* nits

* nits (2)

* fix order

* llama_model_base

* LLAMA_LOAD_LOCALS

* small fix

* fix build errors

* auto

* rm migration script and ifdef
This commit is contained in:
Xuan-Son Nguyen
2026-05-04 12:36:59 +02:00
committed by GitHub
parent c84e6d6db5
commit 994118a183
129 changed files with 10667 additions and 8117 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-quant.cpp
+14 -9
View File
@@ -882,13 +882,18 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
fname_inp, splits, /*file*/ nullptr, use_mmap, /*use_direct_io*/ false, /*check_tensors*/ true, /*no_alloc*/ false, kv_overrides, nullptr);
ml.init_mappings(false); // no prefetching
llama_model model(llama_model_default_params());
auto mparams = llama_model_default_params();
std::unique_ptr<llama_model> model_ptr(llama_model_create(ml, mparams));
model.load_arch (ml);
model.load_hparams(ml);
model.load_stats (ml);
auto * model = dynamic_cast<llama_model_base *>(model_ptr.get());
if (model == nullptr) {
GGML_ABORT("fatal error: model does not implement llama_model_base");
}
quantize_state_impl qs(model, params);
model->load_hparams(ml);
model->load_stats (ml);
quantize_state_impl qs(*model, params);
if (params->only_copy) {
ftype = ml.ftype;
@@ -1023,7 +1028,7 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
}
gguf_add_tensor(ctx_outs[i_split].get(), tensor);
metadata[i].allows_quantization = tensor_allows_quantization(params, model.arch, tensor);
metadata[i].allows_quantization = tensor_allows_quantization(params, model->arch, tensor);
if (metadata[i].allows_quantization) {
metadata[i].target_type = llama_tensor_get_type(qs, params, tensor, default_type, metadata[i]);
@@ -1331,9 +1336,9 @@ void llama_quant_free(quantize_state_impl * qs) {
llama_model * llama_quant_model_from_metadata(const llama_quant_model_desc * desc) {
struct llama_model_params mparams = llama_model_default_params();
auto * model = new llama_model(mparams);
model->arch = llm_arch_from_string(desc->architecture);
auto arch = llm_arch_from_string(desc->architecture);
auto * model = llama_model_create(arch, mparams);
model->arch = arch;
// infer llm_type: only LLM_TYPE_70B matters for quantization logic
if (model->arch == LLM_ARCH_LLAMA && desc->n_layer == 80 && desc->n_head != desc->n_head_kv) {