Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-08 05:49:10 +02:00
Code Issues Packages Projects Releases Wiki Activity

added support for 70b llama 2

Browse Source
This commit is contained in:
Concedo
2023-07-24 15:20:18 +08:00
parent 993ba3b026
commit 94499dba25
1 changed files with 5 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
llama.cpp
+5
View File
@@ -1045,6 +1045,11 @@ static void llama_model_load_internal(
// LLaMAv2
// TODO: temporary until GGUF
//patch for llama2 gqa
if (model.type == e_model::MODEL_65B && hparams.n_mult >= 4096) {
fprintf(stderr, "%s: Applying KCPP Patch for 70B model, setting GQA to 8\n", __func__);
n_gqa = 8;
}
LLAMA_ASSERT(hparams.n_head % n_gqa == 0);
hparams.n_head_kv = hparams.n_head / n_gqa;
if (model.type == e_model::MODEL_65B && n_gqa == 8) {