ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add unittest

Browse Source 
Former-commit-id: 8a1f0c5f922989e08a19c65de0b2c4afd2a5771f
This commit is contained in:
hiyouga
2024-07-19 01:06:27 +08:00
parent 4c1513a845
commit 994b9089e9
16 changed files with 436 additions and 260 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llamafactory/train/ppo/trainer.py
View File

@@ -119,7 +119,9 @@ class CustomPPOTrainer(PPOTrainer, Trainer):
num_training_steps = training_args.max_steps
else:
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(train_dataset) / total_train_batch_size)
num_training_steps = training_args.num_train_epochs * math.ceil(
len(train_dataset) / total_train_batch_size
)
optimizer = self.create_optimizer(model, training_args, finetuning_args)
scheduler = self.create_scheduler(training_args, num_training_steps, optimizer)