Files
pythia410m-sft-tldr/code/configs/gptrm_sft_pythia410m_amp16.yml

14 lines
335 B
YAML
Raw Normal View History

model_name: /home/toolkit/huggingface/tldr_sft_pythia410m_fp32
load_in_8bit: False
bf16: False
fp16: True
use_lora: False
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