Skip to content

Add GRPO micro-batch size validation to prevent incorrect grouping and advantage computation #820

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
aaghaazkhan wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Add GRPO micro-batch size validation to prevent incorrect grouping and advantage computation #820

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d2227af
Fix: enforce GRPO micro-batch grouping validation to prevent incorrec…
aaghaazkhan Dec 1, 2025
6bbafa6
Merge branch 'main' into fix-grpo-microbatch
aaghaazkhan Apr 14, 2026
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
17 changes: 17 additions & 0 deletions tunix/rl/grpo/grpo_learner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -218,6 +218,23 @@ def _generate_and_compute_advantage(
pad_value = self.rl_cluster.rollout.pad_id()
eos_value = self.rl_cluster.rollout.eos_id()

# === GRPO FIX: validate grouping ===
if self._rollout_micro_batch_size < self.algo_config.num_generations:
raise ValueError(
f"GRPO requires rollout_micro_batch_size >= num_generations. "
f"Got rollout_micro_batch_size={self._rollout_micro_batch_size}, "
f"num_generations={self.algo_config.num_generations}. "
f"Increase train_micro_batch_size or reduce num_generations."
)

if self._compute_logps_micro_batch_size < self.algo_config.num_generations:
raise ValueError(
f"GRPO requires compute_logps_micro_batch_size >= num_generations. "
f"Got compute_logps_micro_batch_size={self._compute_logps_micro_batch_size}, "
f"num_generations={self.algo_config.num_generations}. "
f"Increase compute_logps_micro_batch_size or reduce num_generations."
)
# ==================================
# TODO (noghabi): Add mini batch and micro batch tags
perf_tags = {
perf_constants.STEP: self.rl_cluster.global_steps,
Expand Down