Skip to content

Fix fsdp prob diff computation when tis is disabled #82

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

Merged
fzyzcjy merged 1 commit into from
Oct 24, 2025
+3 −3
Merged

Fix fsdp prob diff computation when tis is disabled #82

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
6 changes: 3 additions & 3 deletions miles/backends/fsdp_utils/actor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -430,6 +430,9 @@ def train(self, rollout_id: int, rollout_data_ref: Box) -> None:

pg_loss, pg_clipfrac = compute_policy_loss(ppo_kl, advantages, self.args.eps_clip, self.args.eps_clip_high)

rollout_log_probs = torch.cat([batch["rollout_log_probs"] for batch in unpacked_batches], dim=0)
rollout_log_probs = rollout_log_probs.to(device=log_probs.device)

# Apply TIS before sample mean calculation
if self.args.use_tis:
# Initialize TIS variables
Expand All @@ -444,9 +447,6 @@ def train(self, rollout_id: int, rollout_data_ref: Box) -> None:
for batch in unpacked_batches
), "rollout_log_probs must be provided as non-empty torch.Tensor for TIS"

rollout_log_probs = torch.cat([batch["rollout_log_probs"] for batch in unpacked_batches], dim=0)
rollout_log_probs = rollout_log_probs.to(device=log_probs.device)

tis = torch.exp(old_log_probs - rollout_log_probs)
ois = (-ppo_kl).exp()
tis_clip = torch.clamp(
Expand Down