output_dir: summarize_from_feedback_tldr3_generated_20k_relabel_pythia1b_dpo_temp0.7_length128 mode: relabel model_name: vwxyzjn/EleutherAI_pythia-1b-deduped__dpo__tldr model_revision: dpo__55513__1707379566 ref_model_name: vwxyzjn/EleutherAI_pythia-1b-deduped__sft__tldr ref_model_revision: sft__55513__1706646024 tokenizer_name: EleutherAI/pythia-1b-deduped dataset_name: mnoukhov/summarize_from_feedback_tldr3_generated_20k_vllm_pythia1b_dpo_temp0.7_length128 max_prompt_length: 512 max_target_length: 128 max_length: 640 eval_split: train use_peft: False beta: 0.5 load_in_8bit: False bf16: True fp16: False per_device_eval_batch_size: 8 warmup_steps: 150