ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-10 22:10:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

Force split_mode_f16 to false

Browse Source
This commit is contained in:
Iwan Kawrakow
2026-01-07 14:58:59 +00:00
parent 3cfb1ad6d8
commit 646fe94085
1 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
src/llama.cpp
View File

@@ -4433,6 +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);