| 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/loss/value_avg | train/val/clipfrac_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']}) | 4.7787 ± 0.0000 | 6.5834 ± 0.0000 | 36.0650 ± 0.0000 | -1.8035 ± 0.0000 | 22.1925 ± 0.0000 | 0.0030 ± 0.0000 | 0.0163 ± 0.0000 | 0.0011 ± 0.0000 | 0.0615 ± 0.0000 | 0.0002 ± 0.0000 | 0.6419 ± 0.0000 | 0.9995 ± 0.0000 | 0.0000 ± 0.0000 | 526.8100 ± 0.0000 | 0.0000 ± 0.0000 | 32.0000 ± 0.0000 |