Skip to content

Fix PPO's compute_reward() (#996) #997

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
CZH-THU wants to merge 1 commit into
base: master
Choose a base branch
from
+2 −2
Open

Fix PPO's compute_reward() (#996) #997

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions applications/DeepSpeed-Chat/dschat/rlhf/ppo_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -184,7 +184,7 @@ def compute_rewards(self, prompts, log_probs, ref_log_probs, reward_score,
kl_divergence_estimate = -self.kl_ctl * (log_probs - ref_log_probs)
rewards = kl_divergence_estimate
start = prompts.shape[1] - 1
ends = start + action_mask[:, start:].sum(1) + 1
ends = start + action_mask[:, start:].sum(1)
reward_clip = torch.clamp(reward_score, -self.clip_reward_value,
self.clip_reward_value)
batch_size = log_probs.shape[0]
Expand Down Expand Up @@ -212,7 +212,7 @@ def train_rlhf(self, inputs):
old_rewards = self.compute_rewards(prompts, log_probs,
ref_log_probs, reward_score,
action_mask)
ends = start + action_mask[:, start:].sum(1) + 1
ends = start + action_mask[:, start:].sum(1)
# we need to zero out the reward and value after the end of the conversation
# otherwise the advantage/return will be wrong
for i in range(old_rewards.shape[0]):
Expand Down