mirror of
https://github.com/hanzoai/open-instruct.git
synced 2026-11-03 02:13:07 +00:00
e1e52f5add
* Update valpy_dpo.yaml * Update configs/train_configs/dpo/valpy_dpo.yaml Co-authored-by: Hamish Ivison <[email protected]> --------- Co-authored-by: Hamish Ivison <[email protected]>
23 lines
538 B
YAML
23 lines
538 B
YAML
model_name_or_path: /model
|
|
model_revision: main
|
|
use_flash_attn: true
|
|
gradient_checkpointing: true
|
|
tokenizer_name: /model
|
|
use_slow_tokenizer: true
|
|
dataset_name: valpy/uf_anteater_wildchat
|
|
max_seq_length: 2048
|
|
preprocessing_num_workers: 16
|
|
per_device_train_batch_size: 1
|
|
gradient_accumulation_steps: 4 # designed for 8 GPUs, so batch size 32
|
|
learning_rate: 5.0e-7
|
|
lr_scheduler_type: linear
|
|
warmup_ratio: 0.1
|
|
weight_decay: 0.0
|
|
num_train_epochs: 3
|
|
output_dir: /output
|
|
with_tracking: true
|
|
report_to:
|
|
- wandb
|
|
logging_steps: 1
|
|
use_lora: false
|