Trying grad checkpoint

This commit is contained in:
Jake Poznanski 2024-10-08 16:11:31 +00:00
parent 4fb7e9b184
commit f5fd9ff53a

View File

@ -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