update packing with sdpa and eager attention mode

Former-commit-id: 285636ba3a57a1038b2f2fd4cf909a1ca07708d4
2024-06-16 02:25:47 +07:00
parent f91fe10985
commit dd7a1dbfae
3 changed files with 148 additions and 73 deletions
--- a/src/llamafactory/train/sft/workflow.py
+++ b/src/llamafactory/train/sft/workflow.py
@@ -34,7 +34,7 @@ def run_sft(
    model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train)

    if data_args.efficient_packing:
-        configure_packing(model.config)
+        configure_packing(model.config, model_args)

    if training_args.predict_with_generate:
        tokenizer.padding_side = "left"  # use left-padding in generation