jacobeen06 commited on
Commit
9d44c0c
·
verified ·
1 Parent(s): bff90ea

Skip eval, disable MoE aux loss, disable dropout to fix OOM

Browse files
Files changed (1) hide show
  1. train_dpo.py +3 -1
train_dpo.py CHANGED
@@ -189,13 +189,15 @@ def train_dpo(config: Config):
189
  logging_steps=config.logging_steps,
190
  save_steps=config.save_steps,
191
  eval_steps=config.eval_steps,
192
- eval_strategy="steps" if len(eval_dataset) > 0 else "no",
193
  save_total_limit=config.save_total_limit,
194
  remove_unused_columns=config.remove_unused_columns,
195
  bf16=True,
196
  tf32=True,
197
  gradient_checkpointing=True,
198
  gradient_checkpointing_kwargs={"use_reentrant": False},
 
 
199
  logging_dir=os.path.join(config.output_dir, "logs"),
200
  report_to="wandb" if os.environ.get("WANDB_API_KEY") else "none",
201
  run_name="genesis2-dpo",
 
189
  logging_steps=config.logging_steps,
190
  save_steps=config.save_steps,
191
  eval_steps=config.eval_steps,
192
+ eval_strategy="no", # Skip eval to avoid MoE aux loss OOM
193
  save_total_limit=config.save_total_limit,
194
  remove_unused_columns=config.remove_unused_columns,
195
  bf16=True,
196
  tf32=True,
197
  gradient_checkpointing=True,
198
  gradient_checkpointing_kwargs={"use_reentrant": False},
199
+ disable_dropout=True,
200
+ router_aux_loss_coef=0.0,
201
  logging_dir=os.path.join(config.output_dir, "logs"),
202
  report_to="wandb" if os.environ.get("WANDB_API_KEY") else "none",
203
  run_name="genesis2-dpo",