Remove ignored max_length parameter from PRMTrainer data collator (#4355)

albertvillanova · web-flow · commit d2754185dbe9 · 2025-10-29T11:13:33.000+01:00
diff --git a/trl/trainer/prm_trainer.py b/trl/trainer/prm_trainer.py
@@ -141,7 +141,7 @@ def __init__(
                 raise ValueError(
                     "A processing_class must be specified when using the default DataCollatorForTokenClassification"
                 )
-            data_collator = DataCollatorForTokenClassification(processing_class, max_length=args.max_length)
+            data_collator = DataCollatorForTokenClassification(processing_class)
 
         if "input_ids" not in train_dataset.column_names:
             with PartialState().main_process_first():

Original file line number	Diff line number	Diff line change
`@@ -141,7 +141,7 @@ def __init__(`
`141`	`141`	`raise ValueError(`
`142`	`142`	`"A processing_class must be specified when using the default DataCollatorForTokenClassification"`
`143`	`143`	`)`
`144`		`- data_collator = DataCollatorForTokenClassification(processing_class, max_length=args.max_length)`
	`144`	`+ data_collator = DataCollatorForTokenClassification(processing_class)`
`145`	`145`
`146`	`146`	`if "input_ids" not in train_dataset.column_names:`
`147`	`147`	`with PartialState().main_process_first():`