Disable split mode graph for Qwen35-MoE when MTP is enabled (#1858)
This commit is contained in:
parent
4b73de246b
commit
48a55f74e4
|
|
@ -3319,6 +3319,12 @@ static bool llm_load_tensors(
|
|||
LLAMA_LOG_WARN(" => changing split mode to 'layer'\n");
|
||||
LLAMA_LOG_WARN("===========================================================\n\n");
|
||||
split_mode = LLAMA_SPLIT_MODE_LAYER;
|
||||
} else if (model.arch == LLM_ARCH_QWEN35MOE && mtp) {
|
||||
LLAMA_LOG_WARN("\n=========================================================\n");
|
||||
LLAMA_LOG_WARN("Split mode 'graph' curently does not work with MTP for Qwen35-MoE\n");
|
||||
LLAMA_LOG_WARN(" => changing split mode to 'layer'\n");
|
||||
LLAMA_LOG_WARN("=======================================================\n\n");
|
||||
split_mode = LLAMA_SPLIT_MODE_LAYER;
|
||||
} else if (!is_model_split_supported(model)) {
|
||||
LLAMA_LOG_WARN("\n=======================================================\n");
|
||||
LLAMA_LOG_WARN("Split mode 'graph' is not supported for this model\n");
|
||||
|
|
|
|||
Loading…
Reference in New Issue