Skip to content

feat(docs): save checkpoints natively in a HF format #30

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
merged 2 commits into from
Oct 27, 2025
+4 −26
Merged

feat(docs): save checkpoints natively in a HF format #30

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e5319e9
feat(docs): save checkpoints natively in a HF format
Oct 21, 2025
838b84d
Merge branch 'main' into new-flag
Oct 27, 2025
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
15 changes: 2 additions & 13 deletions examples/qwen3-8b-fft-verl/training/run.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,7 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=True \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=True \
actor_rollout_ref.actor.checkpoint.save_contents=[model,optimizer,extra,hf_model] \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=4 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
actor_rollout_ref.rollout.name=vllm \
Expand All @@ -66,16 +67,4 @@ python3 -m verl.trainer.main_ppo \
algorithm.rollout_is_threshold=2.0 \
algorithm.rollout_is=true \
algorithm.rollout_is_level=token \
algorithm.rollout_is_mode=truncate \

for checkpoint_dir in $BT_CHECKPOINT_DIR/global_step_*/; do
if [ -d "$checkpoint_dir/actor" ]; then
echo "Merging actor model from $(basename $checkpoint_dir)..."
python -m verl.model_merger merge \
--backend fsdp \
--local_dir "$checkpoint_dir/actor" \
--target_dir "$checkpoint_dir/actor_hf"
else
echo "No actor directory found in $(basename $checkpoint_dir), skipping..."
fi
done
algorithm.rollout_is_mode=truncate \
Copy link
Collaborator

Choose a reason for hiding this comment

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

can you help me understand what this does?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

I followed the migration guide here to fix a breaking change in the upstream VERL.
https://github.com/szrlee/verl/blob/d651fce18f2c18836c605320d7d13e09b9e27d23/docs/advance/rollout_is_migration.md

15 changes: 2 additions & 13 deletions examples/qwen3-8b-lora-verl/training/run.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@ python3 -m verl.trainer.main_ppo \
actor_rollout_ref.model.enable_gradient_checkpointing=True \
actor_rollout_ref.actor.fsdp_config.param_offload=False \
actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
actor_rollout_ref.actor.checkpoint.save_contents=[model,optimizer,extra,hf_model] \
actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=1 \
actor_rollout_ref.rollout.tensor_model_parallel_size=2 \
actor_rollout_ref.rollout.name=vllm \
Expand All @@ -57,16 +58,4 @@ python3 -m verl.trainer.main_ppo \
algorithm.rollout_is_threshold=2.0 \
algorithm.rollout_is=true \
algorithm.rollout_is_level=token \
algorithm.rollout_is_mode=truncate \

for checkpoint_dir in $BT_CHECKPOINT_DIR/global_step_*/; do
if [ -d "$checkpoint_dir/actor" ]; then
echo "Merging actor model from $(basename $checkpoint_dir)..."
python -m verl.model_merger merge \
--backend fsdp \
--local_dir "$checkpoint_dir/actor" \
--target_dir "$checkpoint_dir/actor_hf"
else
echo "No actor directory found in $(basename $checkpoint_dir), skipping..."
fi
done
algorithm.rollout_is_mode=truncate \