Files
pythia410m-sft-tldr/code/configs/dpo1b_relabel_generated_pythia410m_fp16.yml

13 lines
347 B
YAML
Raw Normal View History

output_dir: /home/toolkit/huggingface/openai_summarize_generated_20k_relabelled_margin
mode: relabel
model_name: mnoukhov/pythia410m-tldrprompt-dpo1b-adapter
dataset_name: mnoukhov/openai_summarize_generated_20k
eval_split: train
use_peft: False
beta: 0.5
load_in_8bit: False
bf16: False
fp16: True
per_device_eval_batch_size: 8
warmup_steps: 150