fix: mixtral output_router_logits

Former-commit-id: 9f4fe623866b10b30c6418dee116b36671274f9f
This commit is contained in:
liuzc 2024-04-15 12:11:49 +08:00
parent db42378f29
commit 11f4afc5ad

View File

@ -316,7 +316,7 @@ def patch_config(
if getattr(config, "model_type", None) == "qwen2" and is_trainable and model_args.flash_attn:
setattr(config, "use_cache", False) # qwen2 does not support use_cache when using flashattn
if getattr(config, "model_type", None) == "qwen2_moe" and is_trainable:
if getattr(config, "model_type", None) in ["mixtral", "qwen2_moe"] and is_trainable:
setattr(config, "output_router_logits", True)
init_kwargs["torch_dtype"] = model_args.compute_dtype