output_dir: compare_test
# output_dir: openai_summarize_generated_20k_vllm_pythia410m_dpo1
model_name: mnoukhov/pythia410m-tldrprompt-dpo1b
dataset_name: arianhosseini/openai_summarize_unlabelled
train_split: train[:20000]
batch_size: 4
seq_length: 512
max_new_tokens: 48
temperature: 0.7
top_p: 1.0