model_name: mnoukhov/pythia410m-tldr-sft
dataset_name: mnoukhov/openai_summarize_comparisons_tldrprompt_relabel1b
load_in_8bit: False
bf16: False
fp16: True
per_device_train_batch_size: 8
per_device_eval_batch_size: 8
gradient_accumulation_steps: 4
gradient_checkpointing: False
learning_rate: 2e-5
optimizer_type: adamw_torch
num_train_epochs: 1
use_lora: True
lora_all_linear: True
lora_r: 8
lora_alpha: 32
padding: do_not_pad
eval_steps: 0.25