From 32656bc50dc2d51b41a4f2da485fba0070902b2f Mon Sep 17 00:00:00 2001 From: Ting Date: Tue, 19 Nov 2024 19:12:10 +0800 Subject: [PATCH] update Former-commit-id: f566ecc8d1f04615351acbe4f8480b75b2daed42 --- src/llamafactory/train/dpo/workflow.py | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/src/llamafactory/train/dpo/workflow.py b/src/llamafactory/train/dpo/workflow.py index 04f534ce..c0767880 100644 --- a/src/llamafactory/train/dpo/workflow.py +++ b/src/llamafactory/train/dpo/workflow.py @@ -66,10 +66,10 @@ def run_dpo( # Update arguments training_args.remove_unused_columns = False # important for multimodal and pairwise dataset - effi_token_num = 0.0 + effective_token_num = 0.0 for data in dataset_module["train_dataset"]: - effi_token_num += len(data["chosen_input_ids"]) - effi_token_num += len(data["rejected_input_ids"]) + effective_token_num += len(data["chosen_input_ids"]) + effective_token_num += len(data["rejected_input_ids"]) # Initialize our Trainer trainer = CustomDPOTrainer( @@ -87,7 +87,7 @@ def run_dpo( if training_args.do_train: train_result = trainer.train(resume_from_checkpoint=training_args.resume_from_checkpoint) train_result.metrics["effective_tokens_per_sec"] = ( - effi_token_num * train_result.metrics["epoch"] / train_result.metrics["train_runtime"] + effective_token_num * train_result.metrics["epoch"] / train_result.metrics["train_runtime"] ) if dist.is_initialized(): train_result.metrics["effective_tokens_per_sec"] = (