mirror of
https://github.com/wassname/SimPO.git
synced 2026-06-27 16:43:59 +08:00
Yu Meng
46 lines
980 B
YAML
46 lines
980 B
YAML
# Model arguments
|
|
model_name_or_path: princeton-nlp/Llama-3-Base-8B-SFT
|
|
torch_dtype: null
|
|
attn_implementation: flash_attention_2
|
|
|
|
# Data training arguments
|
|
dataset_mixer:
|
|
HuggingFaceH4/ultrafeedback_binarized: 1.0
|
|
dataset_splits:
|
|
- train_prefs
|
|
- test_prefs
|
|
preprocessing_num_workers: 12
|
|
|
|
# SimPOTrainer arguments
|
|
bf16: true
|
|
beta: 2.0
|
|
gamma_beta_ratio: 0.5
|
|
do_eval: true
|
|
evaluation_strategy: steps
|
|
eval_steps: 400
|
|
gradient_accumulation_steps: 16
|
|
gradient_checkpointing: true
|
|
gradient_checkpointing_kwargs:
|
|
use_reentrant: False
|
|
hub_model_id: simpo-exps
|
|
learning_rate: 6.0e-7
|
|
log_level: info
|
|
logging_steps: 5
|
|
lr_scheduler_type: cosine
|
|
max_length: 2048
|
|
max_prompt_length: 1800
|
|
num_train_epochs: 1
|
|
optim: adamw_torch
|
|
output_dir: outputs/llama-3-8b-base-simpo
|
|
run_name: llama-3-8b-base-simpo
|
|
per_device_train_batch_size: 2
|
|
per_device_eval_batch_size: 4
|
|
push_to_hub: false
|
|
save_strategy: "steps"
|
|
save_steps: 1000000
|
|
report_to:
|
|
- wandb
|
|
save_total_limit: 20
|
|
seed: 42
|
|
warmup_ratio: 0.1
|