dataset_name: mnoukhov/openai_summarize_comparisons_relabel_pythia7b
model_name: /home/toolkit/huggingface/tldr_sft_pythia410m_fp32
load_in_8bit: False
bf16: False
fp16: True
per_device_train_batch_size: 8
per_device_eval_batch_size: 8
gradient_accumulation_steps: 16
gradient_checkpointing: False
learning_rate: 1e-5
optimizer_type: paged_adamw_32bit
num_train_epochs: 3
seq_length: 560
use_lora: True
lora_dropout: 0.01
lora_all_linear: True
lora_r: 8