Skip to content

[algo] fix: correctly aggregate kl metrics in PPO actor #2259

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 3 commits into from
Jun 30, 2025
Merged

[algo] fix: correctly aggregate kl metrics in PPO actor #2259

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
8e57fcc
[actor] fix: wrong kl metrics collection in DataParallelPPOActor
0x404 Jun 29, 2025
c5c8f38
[critic] refactor: improve micro-batch metrics collection in DataPara…
0x404 Jun 29, 2025
3d8bf7a
[critic] fix: ensure critic learning rate scheduler is updated correctly
0x404 Jun 29, 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
26 changes: 15 additions & 11 deletions verl/workers/actor/dp_actor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -414,6 +414,8 @@ def update_policy(self, data: DataProto):
self.actor_optimizer.zero_grad()

for data in micro_batches:
micro_batch_metrics = {}

# Support all hardwares
if isinstance(data, DataProto):
data = {**data.batch.to(get_device_id()), **data.non_tensor_batch}
Expand Down Expand Up @@ -484,8 +486,8 @@ def update_policy(self, data: DataProto):
kl_loss = agg_loss(loss_mat=kld, loss_mask=response_mask, loss_agg_mode=loss_agg_mode)

policy_loss = policy_loss + kl_loss * self.config.kl_loss_coef
metrics["actor/kl_loss"] = kl_loss.detach().item()
metrics["actor/kl_coef"] = self.config.kl_loss_coef
micro_batch_metrics["actor/kl_loss"] = kl_loss.detach().item()
micro_batch_metrics["actor/kl_coef"] = self.config.kl_loss_coef

if self.config.use_dynamic_bsz:
# relative to the dynamic bsz
Expand All @@ -494,16 +496,18 @@ def update_policy(self, data: DataProto):
loss = policy_loss / self.gradient_accumulation
loss.backward()

data = {
"actor/pg_loss": pg_loss.detach().item(),
"actor/pg_clipfrac": pg_clipfrac.detach().item(),
"actor/ppo_kl": ppo_kl.detach().item(),
"actor/pg_clipfrac_lower": pg_clipfrac_lower.detach().item(),
}
append_to_dict(metrics, data)
micro_batch_metrics.update(
{
"actor/pg_loss": pg_loss.detach().item(),
"actor/pg_clipfrac": pg_clipfrac.detach().item(),
"actor/ppo_kl": ppo_kl.detach().item(),
"actor/pg_clipfrac_lower": pg_clipfrac_lower.detach().item(),
}
)
append_to_dict(metrics, micro_batch_metrics)

grad_norm = self._optimizer_step()
data = {"actor/grad_norm": grad_norm.detach().item()}
append_to_dict(metrics, data)
mini_batch_metrics = {"actor/grad_norm": grad_norm.detach().item()}
append_to_dict(metrics, mini_batch_metrics)
self.actor_optimizer.zero_grad()
return metrics
20 changes: 12 additions & 8 deletions verl/workers/critic/dp_critic.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -221,6 +221,8 @@ def update_critic(self, data: DataProto):
self.critic_optimizer.zero_grad()

for data in micro_batches:
micro_batch_metrics = {}

# Support all devices
if isinstance(data, DataProto):
data = {**data.batch.to(get_device_id()), **data.non_tensor_batch}
Expand Down Expand Up @@ -254,16 +256,18 @@ def update_critic(self, data: DataProto):

loss.backward()

data = {
"critic/vf_loss": vf_loss.detach().item(),
"critic/vf_clipfrac": vf_clipfrac.detach().item(),
"critic/vpred_mean": masked_mean(vpreds, response_mask).detach().item(),
}
micro_batch_metrics.update(
{
"critic/vf_loss": vf_loss.detach().item(),
"critic/vf_clipfrac": vf_clipfrac.detach().item(),
"critic/vpred_mean": masked_mean(vpreds, response_mask).detach().item(),
}
)

append_to_dict(metrics, data)
append_to_dict(metrics, micro_batch_metrics)

grad_norm = self._optimizer_step()
data = {"critic/grad_norm": grad_norm.detach().item()}
append_to_dict(metrics, data)
mini_batch_metrics = {"critic/grad_norm": grad_norm.detach().item()}
append_to_dict(metrics, mini_batch_metrics)
self.critic_optimizer.zero_grad()
return metrics
2 changes: 1 addition & 1 deletion verl/workers/fsdp_workers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1123,9 +1123,9 @@ def update_critic(self, data: DataProto):
estimated_flops, promised_flops = self.flops_counter.estimate_flops(global_num_tokens, delta_time)
metrics["perf/mfu/critic"] = estimated_flops * self.config.ppo_epochs / promised_flops / self.world_size

self.critic_lr_scheduler.step()
lr = self.critic_lr_scheduler.get_last_lr()[0]
metrics["critic/lr"] = lr
self.critic_lr_scheduler.step()
Comment on lines -1126 to +1128
Copy link
Copy Markdown
Collaborator Author

@0x404 0x404 Jun 29, 2025

Choose a reason for hiding this comment

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

also fix an issue where previous recorded lr is actually the next step's lr instead of the current step. as #1463


output = DataProto(batch=None, meta_info={"metrics": metrics})
output = self.ulysses_sharding_manager.postprocess_data(data=output)
Expand Down
Loading