[data] add eval_on_each_dataset arg (#7912)

src/llamafactory/hparams/data_args.py

@@ -99,6 +99,10 @@ class DataArguments:
         default=0.0,
         metadata={"help": "Size of the validation set, should be an integer or a float in range `[0,1)`."},
     )
+    eval_on_each_dataset: bool = field(
+        default=False,
+        metadata={"help": "Whether or not to evaluate on each dataset separately."},
+    )
     packing: Optional[bool] = field(
         default=None,
         metadata={"help": "Enable sequences packing in training. Will automatically enable in pre-training."},

src/llamafactory/hparams/training_args.py

@@ -64,6 +64,7 @@ class RayArguments:
                 raise ValueError(
                     f"ray_storage_filesystem must be one of ['s3', 'gs', 'gcs'], got {self.ray_storage_filesystem}"
                 )
+
             import pyarrow.fs as fs
 
             if self.ray_storage_filesystem == "s3":