423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-29 10:10:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

[model] switch to gptqmodel (#8108)

Browse Source
This commit is contained in:
hoshi-hiyouga
2025-05-19 22:25:40 +08:00
committed by GitHub
parent 8dc195e4ad
commit 8325087bb3
9 changed files with 78 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
src/llamafactory/model/model_utils/moe.py
View File

@@ -99,27 +99,29 @@ def add_z3_leaf_module(model: "PreTrainedModel") -> None:
def configure_moe(config: "PretrainedConfig", model_args: "ModelArguments", is_trainable: bool) -> None:
if not is_trainable or not model_args.moe_aux_loss_coef:
return
model_type = getattr(config, "model_type", None)
if model_args.moe_aux_loss_coef is not None:
if model_type in [
"dbrx",
"granitemoe",
"jamba",
"jetmoe",
"llama4",
"mixtral",
"olmoe",
"phimoe",
"qwen2_moe",
"qwen3_moe",
]:
setattr(config, "output_router_logits", is_trainable)
if model_type in [
"dbrx",
"granitemoe",
"jamba",
"jetmoe",
"llama4",
"mixtral",
"olmoe",
"phimoe",
"qwen2_moe",
"qwen3_moe",
]:
setattr(config, "output_router_logits", True)
if model_type in ["granitemoe", "jamba", "llama4", "mixtral", "olmoe", "phimoe", "qwen2_moe", "qwen3_moe"]:
setattr(config, "router_aux_loss_coef", model_args.moe_aux_loss_coef)
if model_type in ["granitemoe", "jamba", "llama4", "mixtral", "olmoe", "phimoe", "qwen2_moe", "qwen3_moe"]:
setattr(config, "router_aux_loss_coef", model_args.moe_aux_loss_coef)
elif model_type == "deepseek":
setattr(config, "aux_loss_alpha", model_args.moe_aux_loss_coef)
elif model_type == "deepseek":
setattr(config, "aux_loss_alpha", model_args.moe_aux_loss_coef)
elif model_type == "jetmoe":
setattr(config, "aux_loss_coef", model_args.moe_aux_loss_coef)
elif model_type == "jetmoe":
setattr(config, "aux_loss_coef", model_args.moe_aux_loss_coef)