| train/objective/rlhf_reward | train/objective/scores | train/objective/kl | train/objective/non_score_reward | train/objective/entropy | train/policy/approxkl_avg | train/policy/clipfrac_avg | train/loss/policy_avg | train/policy/entropy_avg | train/val/ratio | train/val/ratio_var | train/val/num_eos_tokens | train/lr | train/eps | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| huggingface/trl/cleanrl/EleutherAI_pythia-1b-deduped__sft__tldr ({'tag': ['pr-1540']}) | 3.8473 ± 0.0000 | 4.8672 ± 0.0000 | 20.3925 ± 0.0000 | -1.0196 ± 0.0000 | 3.0505 ± 0.0000 | 0.5018 ± 0.0000 | 0.3758 ± 0.0000 | 0.1565 ± 0.0000 | 0.5307 ± 0.0000 | 0.9997 ± 0.0000 | 0.0000 ± 0.0000 | 2702.9500 ± 0.0000 | 0.0000 ± 0.0000 | 89.0000 ± 0.0000 |