add_bos: false
checkpointing_steps: null
clip_grad_norm: -1
config_name: null
gradient_accumulation_steps: 8
gradient_checkpointing: false
learning_rate: 2.0e-05
logging_steps: 1
lora_alpha: 16
lora_dropout: 0.1
lora_rank: 64
low_cpu_mem_usage: false
lr_scheduler_type: linear
max_seq_length: 2048
max_train_steps: 738
model_name_or_path: ./hf_llama_models/7B
num_train_epochs: 2
output_dir: output/gpt4_cp_2_7B/
overwrite_cache: false
per_device_train_batch_size: 2
preprocessing_num_workers: 16
reduce_loss: mean
report_to: tensorboard
resume_from_checkpoint: null
seed: null
timeout: 1800
train_file: data/gpt4_cp_2.json
trust_remote_code: false
use_8bit_optimizer: false
use_flash_attn: true
use_lora: false
use_qlora: false
use_slow_tokenizer: true
warmup_ratio: 0.03
weight_decay: 0.0
with_tracking: true