ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-26 01:19:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

Support GigaChat3 (#995)

Browse Source 
* Fixing Gigachat support

* Gigachat: CUDA FA (needs 192 x 192 for MLA = 3)

* Gigachat: CPU FA (needs 192 x 192 for MLA = 3)

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-11-24 06:55:14 +01:00
committed by GitHub
parent f6163dd58f
commit 920f424929
11 changed files with 103 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llama-hparams.cpp
View File

@@ -762,8 +762,10 @@ void llm_load_hparams(
for (auto& item : hparams.n_head_kv_arr) item = n_nead_kv;
hparams.n_embd_head_k = 192;
hparams.n_embd_head_v = 128;
ml.get_key(LLM_KV_ATTENTION_KEY_LENGTH_MLA, hparams.n_embd_head_k);
ml.get_key(LLM_KV_ATTENTION_VALUE_LENGTH_MLA, hparams.n_embd_head_v);
}
bool is_lite = (hparams.n_layer == 27);
bool is_lite = (hparams.n_layer == 27 || hparams.n_layer == 26);
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
ml.get_key(LLM_KV_LEADING_DENSE_BLOCK_COUNT, hparams.n_layer_dense_lead);
if (!is_lite) {