diff --git a/pdelfin/train/config/qwen2vl-7b.yaml b/pdelfin/train/config/qwen2vl-7b.yaml index 49d4842..c9c3209 100644 --- a/pdelfin/train/config/qwen2vl-7b.yaml +++ b/pdelfin/train/config/qwen2vl-7b.yaml @@ -52,7 +52,7 @@ hparams: batch_size: 1 eval_batch_size: 1 gradient_accumulation_steps: 4 - gradient_checkpointing: false + gradient_checkpointing: true clip_grad_norm: 1.0 learning_rate: 3e-4 max_steps: 9000