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

llama : add llama_vocab, functions -> methods, naming (#11110)

Browse source 
* llama : functions -> methods (#11110)

* llama : add struct llama_vocab to the API (#11156)

ggml-ci

* hparams : move vocab params to llama_vocab (#11159)

ggml-ci

* vocab : more pimpl (#11165)

ggml-ci

* vocab : minor tokenization optimizations (#11160)

ggml-ci

Co-authored-by: Diego Devesa <slarengh@gmail.com>

* lora : update API names (#11167)

ggml-ci

* llama : update API names to use correct prefix (#11174)

* llama : update API names to use correct prefix

ggml-ci

* cont

ggml-ci

* cont

ggml-ci

* minor [no ci]

* vocab : llama_vocab_add_[be]os -> llama_vocab_get_add_[be]os (#11174)

ggml-ci

* vocab : llama_vocab_n_vocab -> llama_vocab_n_tokens (#11174)

ggml-ci

---------

Co-authored-by: Diego Devesa <slarengh@gmail.com>
This commit is contained in:
Georgi Gerganov 2025-01-12 11:32:42 +02:00 • committed by GitHub
parent c05e8c9934
commit afa8a9ec9b
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
68 changed files with 5855 additions and 5400 deletions
Download patch file Download diff file Expand all files Collapse all files

14
src/llama-quant.cpp
View file

@ -235,7 +235,7 @@ static ggml_type llama_tensor_get_type(quantize_state_impl & qs, ggml_type new_t
else if ((ftype == LLAMA_FTYPE_MOSTLY_Q4_K_M || ftype == LLAMA_FTYPE_MOSTLY_Q5_K_M) &&
use_more_bits(qs.i_attention_wv, qs.n_attention_wv)) new_type = GGML_TYPE_Q6_K;
else if (ftype == LLAMA_FTYPE_MOSTLY_Q4_K_S && qs.i_attention_wv < 4) new_type = GGML_TYPE_Q5_K;
if (qs.model.type == MODEL_70B) {
if (qs.model.type == LLM_TYPE_70B) {
// In the 70B model we have 8 heads sharing the same attn_v weights. As a result, the attn_v.weight tensor is
// 8x smaller compared to attn_q.weight. Hence, we can get a nice boost in quantization accuracy with
// nearly negligible increase in model size by quantizing this tensor with more bits:
@ -525,18 +525,20 @@ static void llama_model_quantize_impl(const std::string & fname_inp, const std::
auto v = (std::vector<llama_model_kv_override>*)params->kv_overrides;
kv_overrides = v->data();
}
llama_model_loader ml(fname_inp, use_mmap, /*check_tensors*/ true, kv_overrides);
ml.init_mappings(false); // no prefetching
llama_model model;
llm_load_arch (ml, model);
llm_load_hparams(ml, model);
llm_load_stats (ml, model);
llama_model model(llama_model_default_params());
model.load_arch (ml);
model.load_hparams(ml);
model.load_stats (ml);
struct quantize_state_impl qs(model, params);
if (params->only_copy) {
ftype = model.ftype;
ftype = ml.ftype;
}
const std::unordered_map<std::string, std::vector<float>> * imatrix_data = nullptr;
if (params->imatrix) {