[trainer] Add Muon Optimizer (#7749)

Co-authored-by: hoshi-hiyouga <hiyouga@buaa.edu.cn>

src/llamafactory/hparams/finetuning_args.py

@@ -411,6 +411,10 @@ class FinetuningArguments(
         default=False,
         metadata={"help": "Whether or not to use the Adam-mini optimizer."},
     )
+    use_muon: bool = field(
+        default=False,
+        metadata={"help": "Whether or not to use the Muon optimizer."},
+    )
     freeze_vision_tower: bool = field(
         default=True,
         metadata={"help": "Whether ot not to freeze the vision tower in MLLM training."},

src/llamafactory/hparams/parser.py

@@ -153,7 +153,7 @@ def _check_extra_dependencies(
     elif model_args.infer_backend == EngineName.SGLANG:
         check_version("sglang>=0.4.4")
         check_version("sglang", mandatory=True)
-
+    
     if finetuning_args.use_galore:
         check_version("galore_torch", mandatory=True)