ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-30 03:11:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

Remove scary warning about incompatible model (#717)

Browse Source 
* Remove scary warning about incompatible model

* Minor

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-08-22 18:42:01 +03:00
committed by GitHub
parent e962ce8c70
commit 866145b2b9
1 changed files with 6 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
src/llama.cpp
View File

@@ -4072,23 +4072,21 @@ static void llm_load_hparams(
case LLM_ARCH_DEEPSEEK2: case LLM_ARCH_DEEPSEEK2:
{ {
if (hparams.n_head_kv() == 1) { if (hparams.n_head_kv() == 1) {
printf("==========================================================================\n");
printf("Detected incompatible DeepSeek model.\n");
printf("Will try to fix, but there are no guarantees\n\n");
printf("*** Your prompt processing speed will be crippled ***\n\n");
printf("Consider making your own ik_llama.cpp compatible model or\n");
printf("ask the model provider to make one for you,\n");
int n_nead_kv = hparams.n_gqa(); int n_nead_kv = hparams.n_gqa();
if (n_nead_kv%16 != 0 || hparams.n_embd_head_k != 576 || hparams.n_embd_head_v != 512 || if (n_nead_kv%16 != 0 || hparams.n_embd_head_k != 576 || hparams.n_embd_head_v != 512 ||
hparams.n_rot != 64) { hparams.n_rot != 64) {
printf("==========================================================================\n");
printf("Detected incompatible DeepSeek model without a known way to fixc it.\n");
printf("Consider making your own ik_llama.cpp compatible model or\n");
printf("ask the model provider to make one for you,\n\n");
printf("Sorry, uknown model => cannot fix it => bailing out\n"); printf("Sorry, uknown model => cannot fix it => bailing out\n");
printf("==========================================================================\n");
GGML_ABORT("Fatal error"); GGML_ABORT("Fatal error");
} }
printf("================= Adjusted mainline llama.cpp MLA tensors to ik_llama.cpp\n");
for (auto& item : hparams.n_head_kv_arr) item = n_nead_kv; for (auto& item : hparams.n_head_kv_arr) item = n_nead_kv;
hparams.n_embd_head_k = 192; hparams.n_embd_head_k = 192;
hparams.n_embd_head_v = 128; hparams.n_embd_head_v = 128;
printf("==========================================================================\n");
//GGML_ABORT("Fatal error");
} }
bool is_lite = (hparams.n_layer == 27); bool is_lite = (hparams.n_layer == 27);
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps); ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);