try lower lr

This commit is contained in:
Jake Poznanski 2024-10-08 23:52:56 +00:00
parent 230c8a9f9a
commit ad10add6c1

View File

@ -54,8 +54,8 @@ hparams:
gradient_accumulation_steps: 4 gradient_accumulation_steps: 4
gradient_checkpointing: false gradient_checkpointing: false
clip_grad_norm: 1.0 clip_grad_norm: 1.0
learning_rate: 3e-4 learning_rate: 1e-4
max_steps: 9000 max_steps: 20000
pad_multiple_of: 16 pad_multiple_of: 16
log_every_steps: 10 log_every_steps: 10
eval_every_steps: 100 eval_every_steps: 100