File size: 550 Bytes
00db46c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 |
# Default training configuration for GRPO
# Learning parameters
learning_rate: 1e-5
weight_decay: 0.0
warmup_ratio: 0.1
lr_scheduler_type: cosine
optim: adamw_torch
# Training parameters
num_train_epochs: 1
per_device_train_batch_size: 16
gradient_accumulation_steps: 16
bf16: true
remove_unused_columns: false
# Generation parameters
temperature: 1.0
max_completion_length: 256
num_generations: 2
max_prompt_length: 4096
# Logging and saving
report_to: tensorboard
logging_steps: 1
save_strategy: steps
save_steps: 10
|