add_bos: true
checkpointing_steps: epoch
clip_grad_norm: -1
config_name: null
dataset_config_name: null
dataset_name: argilla/ultrafeedback-binarized-preferences-cleaned
dpo_beta: 0.1
dpo_use_paged_optimizer: false
gradient_accumulation_steps: 4
gradient_checkpointing: true
learning_rate: 5.0e-07
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: 4096
max_train_samples: null
max_train_steps: 5712
model_name_or_path: /model
model_revision: main
num_train_epochs: 3
output_dir: /output
overwrite_cache: false
per_device_train_batch_size: 1
preprocessing_num_workers: 16
reduce_loss: mean
report_to: tensorboard
resume_from_checkpoint: null
seed: 42
timeout: 1800
tokenizer_name: /model
tokenizer_revision: main
train_file: null
trust_remote_code: false
use_8bit_optimizer: false
use_flash_attn: true
use_lora: false
use_qlora: false
use_slow_tokenizer: true
wandb_entity: null
warmup_ratio: 0.1
weight_decay: 0.0
with_tracking: true