ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-08 15:30:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Split mode "graph" for GPT-OSS (#1118)

Browse Source 
* Split mode "graph" for GPT-OSS

* Force split_mode_f16 to false

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2026-01-08 09:14:15 +02:00
committed by GitHub
parent 1b24192873
commit d581d75537
3 changed files with 57 additions and 80 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/llama.cpp
View File

@@ -1734,6 +1734,7 @@ static bool is_model_split_supported(const llama_model & model) {
LLM_ARCH_QWEN3,
LLM_ARCH_QWEN3VL,
LLM_ARCH_HUNYUAN_MOE,
LLM_ARCH_OPENAI_MOE,
};
auto it = k_supported.find(model.arch);
return it != k_supported.end();
@@ -4432,6 +4433,15 @@ struct llama_context * llama_new_context_with_model(
//LLAMA_LOG_WARN("=====================================================================\n");
cparams.mla_attn = 0;
}
if (model->arch == LLM_ARCH_OPENAI_MOE && model->split_mode == LLAMA_SPLIT_MODE_GRAPH) {
if (cparams.split_mode_f16) {
LLAMA_LOG_WARN("=====================================================================\n");
LLAMA_LOG_WARN("GPT-OSS with split mode graph requires f32 precision\n");
LLAMA_LOG_WARN(" => changing cparams.split_mode_f16 to 'false'\n");
LLAMA_LOG_WARN("=====================================================================\n");
cparams.split_mode_f16 = false;
}
}
LLAMA_LOG_INFO("%s: n_ctx = %u\n", __func__, cparams.n_ctx);
LLAMA_LOG_INFO("%s: n_batch = %u\n", __func__, cparams.n_batch);