frozenleaves/verl
1
0
Fork 0
mirror of https://github.com/volcengine/verl.git synced 2025-10-20 13:43:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

[trainer] fix: address serialization issues when using async reward function and ray ppo trainer (#3769)

Browse Source
This commit is contained in:
ben
2025-10-17 17:22:59 -07:00
committed by GitHub
parent e0e352b566
commit f0539a5121
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
verl/trainer/ppo/ray_trainer.py
View File

@ -1105,7 +1105,9 @@ class RayPPOTrainer:
batch = batch.union(reward_tensor)
if self.config.reward_model.launch_reward_fn_async:
future_reward = compute_reward_async.remote(data=batch, reward_fn=self.reward_fn)
future_reward = compute_reward_async.remote(
data=batch, config=self.config, tokenizer=self.tokenizer
)
else:
reward_tensor, reward_extra_infos_dict = compute_reward(batch, self.reward_fn)