423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-17 12:20:37 +08:00
Code Issues Packages Projects Releases Wiki Activity

update ppo trainer

Browse Source
This commit is contained in:
hiyouga
2023-11-20 21:39:15 +08:00
parent 48211e3799
commit 5021062493
7 changed files with 68 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/llmtuner/train/ppo/workflow.py
View File

@@ -39,11 +39,12 @@ def run_ppo(
reward_model = create_reward_model(model, model_args, finetuning_args)
# Create ppo config
backward_batch_size = training_args.per_device_train_batch_size * training_args.gradient_accumulation_steps
ppo_config = PPOConfig(
model_name=model_args.model_name_or_path,
learning_rate=training_args.learning_rate,
mini_batch_size=training_args.per_device_train_batch_size,
batch_size=training_args.per_device_train_batch_size * training_args.gradient_accumulation_steps,
batch_size=backward_batch_size * finetuning_args.ppo_buffer_size,
gradient_accumulation_steps=training_args.gradient_accumulation_steps,
ppo_epochs=finetuning_args.ppo_epochs,
max_grad_norm=training_args.max_grad_norm,
@@ -62,9 +63,7 @@ def run_ppo(
if training_args.max_steps > 0:
num_training_steps = training_args.max_steps
else:
total_train_batch_size = (
training_args.per_device_train_batch_size * training_args.gradient_accumulation_steps * training_args.world_size
)
total_train_batch_size = backward_batch_size * finetuning_args.ppo_buffer_size * training_args.world_size
num_training_steps = training_args.num_train_epochs * math.ceil(len(dataset) / total_train_batch_size)
lr_scheduler = get_scheduler(