[recipe, megatron, fsdp] fix: checkpoint-engine fix trainer param offload in fully-async mode (verl-project#4655)

… fully-async mode

### What does this PR do?

> Add **concise** overview of what this PR aims to achieve or
accomplish. Reference related GitHub issues and PRs that help with the
review.

### Checklist Before Starting

- [x] Search for similar PRs. Paste at least one query link here: ...
- [x] Format the PR title as `[{modules}] {type}: {description}` (This
will be checked by the CI)
- `{modules}` include `fsdp`, `megatron`, `sglang`, `vllm`, `rollout`,
`trainer`, `ci`, `training_utils`, `recipe`, `hardware`, `deployment`,
`ray`, `worker`, `single_controller`, `misc`, `perf`, `model`, `algo`,
`env`, `tool`, `ckpt`, `doc`, `data`, `cfg`, `reward`
- If this PR involves multiple modules, separate them with `,` like
`[megatron, fsdp, doc]`
  - `{type}` is in `feat`, `fix`, `refactor`, `chore`, `test`
- If this PR breaks any API (CLI arguments, config, function signature,
etc.), add `[BREAKING]` to the beginning of the title.
  - Example: `[BREAKING][fsdp, megatron] feat: dynamic batching`

### Test

> For changes that can not be tested by CI (e.g., algorithm
implementation, new model support), validate by experiment(s) and show
results like training curve plots, evaluation results, etc.

### API and Usage Example

> Demonstrate how the API changes if any, and provide usage example(s)
if possible.

```python
# Add code snippet or script demonstrating how to use this
```

### Design & Code Changes

> Demonstrate the high-level design if this PR is complex, and list the
specific changes.

### Checklist Before Submitting

> [!IMPORTANT]
> Please check all the following items before requesting a review,
otherwise the reviewer might deprioritize this PR for review.

- [x] Read the [Contribute
Guide](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md).
- [x] Apply [pre-commit
checks](https://github.com/volcengine/verl/blob/main/CONTRIBUTING.md#code-linting-and-formatting):
`pre-commit install && pre-commit run --all-files --show-diff-on-failure
--color=always`
- [ ] Add / Update [the
documentation](https://github.com/volcengine/verl/tree/main/docs).
- [ ] Add unit or end-to-end test(s) to [the CI
workflow](https://github.com/volcengine/verl/tree/main/.github/workflows)
to cover all the code. If not feasible, explain why: ...
- [ ] Once your PR is ready for CI, send a message in [the `ci-request`
channel](https://verl-project.slack.com/archives/C091TCESWB1) in [the
`verl` Slack
workspace](https://join.slack.com/t/verl-project/shared_invite/zt-3855yhg8g-CTkqXu~hKojPCmo7k_yXTQ).
(If not accessible, please try [the Feishu group
(飞书群)](https://applink.larkoffice.com/client/chat/chatter/add_by_link?link_token=772jd4f1-cd91-441e-a820-498c6614126a).)

Author: zpltys

recipe/fully_async_policy/fsdp_workers.py

@@ -135,8 +135,11 @@ def sync_rollout_weights_by_checkpoint(self, sync_group_name="actor_rollout"):
         assert (self._is_actor or self._is_rollout) and not self.config.hybrid_engine
         assert hasattr(self, "_weights_info") and self._weights_info is not None
 
+        # Load model to GPU
+        load_start_time = time.time()
         if self._is_actor and self._is_offload_param:
             load_fsdp_model_to_gpu(self.actor_module_fsdp)
+        load_duration = time.time() - load_start_time
 
         from ray.util.collective import collective
 
@@ -172,14 +175,24 @@ def sync_rollout_weights_by_checkpoint(self, sync_group_name="actor_rollout"):
         update_end_time = time.time()
         update_duration = update_end_time - update_start_time
 
-        collective.barrier(group_name=sync_group_name)
+        offload_start_time = time.time()
+        if self._is_actor and self._is_offload_param:
+            offload_fsdp_model_to_cpu(self.actor_module_fsdp)
+        offload_duration = time.time() - offload_start_time
+
         print(
             f"sync_rollout_weights_by_checkpoint finish!, rank:{torch.distributed.get_rank()},"
             f" is_actor:{self._is_actor}, is_rollout:{self._is_rollout},"
             f" total cost:{update_end_time - cache_start_time} seconds, while cache cost {cache_duration} seconds, "
             f" register cost {register_duration} seconds, update cost {update_duration} seconds"
         )
 
+        if self._is_actor and self._is_offload_param:
+            print(
+                f"sync_rollout_weights_by_checkpoint load model to gpu cost {load_duration} seconds,"
+                f" offload model to cpu cost {offload_duration} seconds"
+            )
+
 
 class DetachActorWorker(DetachNcclSync):
     def _get_actor_params(self):

recipe/fully_async_policy/megatron_worker.py

@@ -140,6 +140,12 @@ def sync_rollout_weights_by_checkpoint(self, sync_group_name="actor_rollout"):
         assert (self._is_actor or self._is_rollout) and not self.config.hybrid_engine
         assert hasattr(self, "_weights_info") and self._weights_info is not None
 
+        # Load model to GPU
+        load_start_time = time.time()
+        if self._is_actor and self._is_offload_param:
+            load_megatron_model_to_gpu(self.actor_module)
+        load_duration = time.time() - load_start_time
+
         from ray.util.collective import collective
 
         # Cache actor weights to CPU and measure the time taken
@@ -174,13 +180,24 @@ def sync_rollout_weights_by_checkpoint(self, sync_group_name="actor_rollout"):
         update_end_time = time.time()
         update_duration = update_end_time - update_start_time
 
+        offload_start_time = time.time()
+        if self._is_actor and self._is_offload_param:
+            offload_megatron_model_to_cpu(self.actor_module)
+        offload_duration = time.time() - offload_start_time
+
         print(
             f"sync_rollout_weights_by_checkpoint finish!, rank:{torch.distributed.get_rank()},"
             f" is_actor:{self._is_actor}, is_rollout:{self._is_rollout},"
             f" total cost:{update_end_time - cache_start_time} seconds, while cache cost {cache_duration} seconds, "
             f" register cost {register_duration} seconds, update cost {update_duration} seconds"
         )
 
+        if self._is_actor and self._is_offload_param:
+            print(
+                f"sync_rollout_weights_by_checkpoint load model to gpu cost {load_duration} seconds,"
+                f" offload model to cpu cost {offload_duration} seconds"
+            )
+
 
 class DetachActorWorker(DetachNcclSync):
     def _get_actor_params_generator(self):