Files
pythia410m-sft-tldr/code/configs/dpo1b_predict_generated_pythia410m-dpo1.yml

12 lines
358 B
YAML
Raw Normal View History

output_dir: /home/toolkit/huggingface/openai_summarize_generated_20k_relabel_1b_predict_410m-dpo1
mode: predict
model_name: mnoukhov/pythia410m-tldrprompt-dpo1b-adapter
dataset_name: mnoukhov/openai_summarize_generated_20k_relabel_1b_margin
eval_split: train
use_peft: False
beta: 0.5
load_in_8bit: False
bf16: False
fp16: True
per_device_eval_batch_size: 8