Skip to content

[doc] fix: set use_dist_checkpointing to False for ref model in qwen3moe-30b script #3198

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
vermouth1992 merged 1 commit into from
Aug 25, 2025
Merged

[doc] fix: set use_dist_checkpointing to False for ref model in qwen3moe-30b script #3198

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 examples/grpo_trainer/run_qwen3moe-30b_megatron_96gb.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -168,7 +168,7 @@ python3 -m verl.trainer.main_ppo --config-path=./config --config-name='ppo_megat
actor_rollout_ref.rollout.free_cache_engine=True \
actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=${infer_ppo_micro_batch_size_per_gpu} \
actor_rollout_ref.ref.log_prob_max_token_len_per_gpu=${infer_ppo_max_token_len} \
actor_rollout_ref.ref.megatron.use_dist_checkpointing=True \
actor_rollout_ref.ref.megatron.use_dist_checkpointing=${USE_DIST_CKPT} \
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot Aug 25, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

critical

Based on the pull request description, use_dist_checkpointing must be False for the reference model because it lacks a distributed checkpoint path. Using the ${USE_DIST_CKPT} variable makes this setting configurable. If a user sets USE_DIST_CKPT=True (e.g., for the actor model), it would also be incorrectly enabled for the reference model, likely causing a runtime error. To ensure the script's robustness and prevent misconfiguration, this value should be hardcoded to False for the reference model.

Suggested change
actor_rollout_ref.ref.megatron.use_dist_checkpointing=${USE_DIST_CKPT} \
actor_rollout_ref.ref.megatron.use_dist_checkpointing=False \

actor_rollout_ref.ref.megatron.param_offload=${offload} \
actor_rollout_ref.ref.megatron.tensor_model_parallel_size=${REF_TP} \
actor_rollout_ref.ref.megatron.pipeline_model_parallel_size=${REF_PP} \
Expand All @@ -192,4 +192,4 @@ python3 -m verl.trainer.main_ppo --config-path=./config --config-name='ppo_megat
trainer.save_freq=100 \
trainer.total_epochs=10 \
trainer.resume_mode=auto \
trainer.log_val_generations=10
trainer.log_val_generations=10