bf16: 'True'
chosen: chosen
cutoff_len: '2048'
dataset: mlfoundations-dev/gemma2-ultrafeedback-armorm
dataset_dir: ONLINE
ddp_timeout: '180000000'
deepspeed: /opt/ml/code/zero3.json
do_train: 'True'
enable_liger_kernel: 'False'
eval_strategy: epoch
finetuning_type: full
formatting: sharegpt
global_batch_size: '128'
gradient_accumulation_steps: '8'
gradient_checkpointing: 'True'
hub_model_id: mlfoundations-dev/simpo-evol_tt_5s
learning_rate: 8e-07
logging_steps: '1'
lr_scheduler_type: cosine
messages: conversations
model_name_or_path: mlfoundations-dev/evol_tt_5s
num_train_epochs: '1.0'
optim: adamw_torch
output_dir: /opt/ml/model
overwrite_cache: 'True'
overwrite_output_dir: 'True'
per_device_eval_batch_size: '2'
per_device_train_batch_size: '2'
plot_loss: 'True'
pref_beta: '10'
pref_loss: simpo
preprocessing_num_workers: '16'
push_to_db: 'True'
push_to_hub: 'True'
ranking: 'True'
rejected: rejected
report_to: wandb
run_name: simpo-evol_tt_5s
save_strategy: epoch
simpo_gamma: '5'
stage: dpo
template: gemma
val_size: '0.05'
warmup_ratio: '0.1'