Skip to main content

Regression Report: wandb

[['?we=costa-huang&wpn=trl&xaxis=_step&ceik=trl_ppo_trainer_config.value.tracker_project_name&cen=trl_ppo_trainer_config.value.log_with&metrics=env/reward_mean&metrics=objective/kl', 'wandb?tag=gpt2-sentiment&tag=rlops-pilot&cl=sentiment analysis (PR-410)', 'wandb?tag=gpt2-sentiment&tag=pr-423&cl=sentiment analysis (PR-423)', 'wandb?tag=gpt2-sentiment-1-nminibs&cl=sentiment analysis (no minibatches, target kl=6)']]
Created on June 28|Last edited on June 28

050100150Steps0.511.52Episodic Return
050100150Steps-60-40-200Episodic Return
51015Time (minutes)0.511.52Episodic Return
51015Time (minutes)-80-60-40-200Episodic Return
sentiment analysis (PR-410)
10
sentiment analysis (PR-423)
9
sentiment analysis (no minibatches, target kl=6)
10