jdelony/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-06-28 04:30:15 -05:00
Code Issues Packages Projects Releases Wiki Activity

It is actually not related to split mode graph

Browse Source
This commit is contained in:
Kawrakow 2026-05-23 10:47:57 +00:00
parent 2c893906e2
commit d065b9f742
1 changed files with 2 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/llama.cpp
View File

@ -3425,9 +3425,7 @@ static bool llm_load_tensors(
}
if (model.arch == LLM_ARCH_DEEPSEEK2 || model.arch == LLM_ARCH_GLM_DSA || model.arch == LLM_ARCH_MISTRAL4) {
if (model.split_mode == LLAMA_SPLIT_MODE_GRAPH &&
model.n_gpu_layers > 0 && model.n_gpu_layers < model.hparams.n_layer &&
mla_attn != 3) {
if (model.n_gpu_layers > 0 && model.n_gpu_layers < model.hparams.n_layer && mla_attn != 3) {
LLAMA_LOG_WARN("=============================================================================\n");
LLAMA_LOG_WARN("MLA models with ngl < n_layer and split mode graph do not work with mla = %d\n", mla_attn);
LLAMA_LOG_WARN(" => changing mla to 3\n");
@ -6762,9 +6760,7 @@ struct llama_context * llama_init_from_model(
if (model->arch != LLM_ARCH_DEEPSEEK2 && model->arch != LLM_ARCH_GLM_DSA && model->arch != LLM_ARCH_MISTRAL4 && cparams.mla_attn != 0) {
cparams.mla_attn = 0;
} else {
if (model->split_mode == LLAMA_SPLIT_MODE_GRAPH &&
model->n_gpu_layers > 0 && model->n_gpu_layers < model->hparams.n_layer &&
cparams.mla_attn != 3) {
if (model->n_gpu_layers > 0 && model->n_gpu_layers < model->hparams.n_layer && cparams.mla_attn != 3) {
LLAMA_LOG_WARN("=============================================================================\n");
LLAMA_LOG_WARN("MLA models with ngl < n_layer and split mode graph do not work with mla = %d\n", cparams.mla_attn);
LLAMA_LOG_WARN(" => changing mla to 3\n");