[trainer] fix: Allow FSDP2 when doing strategy check (#2497)

### What does this PR do?

Allow FSDP2 when doing strategy check

### 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`
- 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.

For `strategy` field, now both "fsdp" and "fsdp2" are considered valid.

### 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`
- [X] Add / Update [the
documentation](https://github.com/volcengine/verl/tree/main/docs).
- [X] 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: ...
- [X] 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).

Signed-off-by: Hollow Man <hollowman@opensuse.org>

docs/examples/ppo_code_architecture.rst

@@ -48,8 +48,8 @@ Define worker classes
 
 .. code:: python
 
-   if config.actor_rollout_ref.actor.strategy == 'fsdp': # for FSDP backend
-       assert config.actor_rollout_ref.actor.strategy == config.critic.strategy
+   if config.actor_rollout_ref.actor.strategy in {"fsdp", "fsdp2"}: # for FSDP backend
+       assert config.critic.strategy in {"fsdp", "fsdp2"}
        from verl.workers.fsdp_workers import ActorRolloutRefWorker, CriticWorker
        from verl.single_controller.ray import RayWorkerGroup
        ray_worker_group_cls = RayWorkerGroup

examples/split_placement/main_ppo_split.py

@@ -123,8 +123,8 @@ def main_task(config):
     tokenizer = hf_tokenizer(local_path)
 
     # define worker classes
-    if config.actor_rollout_ref.actor.strategy == "fsdp":
-        assert config.actor_rollout_ref.actor.strategy == config.critic.strategy
+    if config.actor_rollout_ref.actor.strategy in {"fsdp", "fsdp2"}:
+        assert config.critic.strategy in {"fsdp", "fsdp2"}
         from verl.single_controller.ray import RayWorkerGroup
         from verl.workers.fsdp_workers import ActorRolloutRefWorker, CriticWorker
 
@@ -178,7 +178,7 @@ def main_task(config):
     # - finally, we combine all the rewards together
     # - The reward type depends on the tag of the data
     if config.reward_model.enable:
-        if config.reward_model.strategy == "fsdp":
+        if config.reward_model.strategy in {"fsdp", "fsdp2"}:
             from verl.workers.fsdp_workers import RewardModelWorker
         elif config.reward_model.strategy == "megatron":
             from verl.workers.megatron_workers import RewardModelWorker

recipe/dapo/main_dapo.py

@@ -80,8 +80,8 @@ class TaskRunner:
         processor = hf_processor(local_path, use_fast=True)  # used for multimodal LLM, could be none
 
         # define worker classes
-        if config.actor_rollout_ref.actor.strategy == "fsdp":
-            assert config.actor_rollout_ref.actor.strategy == config.critic.strategy
+        if config.actor_rollout_ref.actor.strategy in {"fsdp", "fsdp2"}:
+            assert config.critic.strategy in {"fsdp", "fsdp2"}
             from verl.single_controller.ray import RayWorkerGroup
             from verl.workers.fsdp_workers import ActorRolloutRefWorker, CriticWorker
 
@@ -120,7 +120,7 @@ class TaskRunner:
         # - finally, we combine all the rewards together
         # - The reward type depends on the tag of the data
         if config.reward_model.enable:
-            if config.reward_model.strategy == "fsdp":
+            if config.reward_model.strategy in {"fsdp", "fsdp2"}:
                 from verl.workers.fsdp_workers import RewardModelWorker
             elif config.reward_model.strategy == "megatron":
                 from verl.workers.megatron_workers import RewardModelWorker

recipe/entropy/main_entropy.py

@@ -85,8 +85,8 @@ class TaskRunner:
         processor = hf_processor(local_path, use_fast=True)  # used for multimodal LLM, could be none
 
         # define worker classes
-        if config.actor_rollout_ref.actor.strategy in ["fsdp", "fsdp2"]:
-            assert config.critic.strategy in ["fsdp", "fsdp2"]
+        if config.actor_rollout_ref.actor.strategy in {"fsdp", "fsdp2"}:
+            assert config.critic.strategy in {"fsdp", "fsdp2"}
             from verl.single_controller.ray import RayWorkerGroup
             from verl.workers.fsdp_workers import ActorRolloutRefWorker, AsyncActorRolloutRefWorker, CriticWorker
 
@@ -131,7 +131,7 @@ class TaskRunner:
         # - finally, we combine all the rewards together
         # - The reward type depends on the tag of the data
         if config.reward_model.enable:
-            if config.reward_model.strategy in ["fsdp", "fsdp2"]:
+            if config.reward_model.strategy in {"fsdp", "fsdp2"}:
                 from verl.workers.fsdp_workers import RewardModelWorker
             elif config.reward_model.strategy == "megatron":
                 from verl.workers.megatron_workers import RewardModelWorker

recipe/prime/main_prime.py

@@ -72,8 +72,8 @@ def main_task(config, compute_score=None):
     tokenizer = hf_tokenizer(local_path)
 
     # define worker classes
-    if config.actor_rollout_ref.actor.strategy == "fsdp":
-        assert config.actor_rollout_ref.actor.strategy == config.critic.strategy
+    if config.actor_rollout_ref.actor.strategy in {"fsdp", "fsdp2"}:
+        assert config.critic.strategy in {"fsdp", "fsdp2"}
         from verl.single_controller.ray import RayWorkerGroup
         from verl.workers.fsdp_workers import ActorRolloutRefWorker
 

recipe/spin/main_spin.py

@@ -67,8 +67,8 @@ class TaskRunner:
         processor = hf_processor(local_path, use_fast=True)  # used for multimodal LLM, could be none
 
         # define worker classes
-        if config.actor_rollout_ref.actor.strategy == "fsdp":
-            assert config.actor_rollout_ref.actor.strategy == config.critic.strategy
+        if config.actor_rollout_ref.actor.strategy in {"fsdp", "fsdp2"}:
+            assert config.critic.strategy in {"fsdp", "fsdp2"}
             # from recipe.spin.fsdp_workers import ActorRolloutRefWorker
             from recipe.spin.fsdp_workers import SPINRolloutRefWorker
             from verl.single_controller.ray import RayWorkerGroup
@@ -102,7 +102,7 @@ class TaskRunner:
         }
 
         if config.reward_model.enable:
-            if config.reward_model.strategy == "fsdp":
+            if config.reward_model.strategy in {"fsdp", "fsdp2"}:
                 from recipe.spin.fsdp_workers import RewardModelWorker
             elif config.reward_model.strategy == "megatron":
                 from verl.workers.megatron_workers import RewardModelWorker

recipe/spin/spin_trainer.py

@@ -522,7 +522,7 @@ class RaySPINTrainer:
                 assert config.critic.ppo_micro_batch_size * sp_size >= n_gpus
 
         # Check if use_remove_padding is enabled when using sequence parallelism for fsdp
-        if config.actor_rollout_ref.actor.strategy == "fsdp":
+        if config.actor_rollout_ref.actor.strategy in {"fsdp", "fsdp2"}:
             if (
                 config.actor_rollout_ref.actor.get("ulysses_sequence_parallel_size", 1) > 1
                 or config.actor_rollout_ref.ref.get("ulysses_sequence_parallel_size", 1) > 1
@@ -531,7 +531,7 @@ class RaySPINTrainer:
                     "When using sequence parallelism for actor/ref policy, you must enable `use_remove_padding`."
                 )
 
-        if self.use_critic and config.critic.strategy == "fsdp":
+        if self.use_critic and config.critic.strategy in {"fsdp", "fsdp2"}:
             if config.critic.get("ulysses_sequence_parallel_size", 1) > 1:
                 assert config.critic.model.use_remove_padding, (
                     "When using sequence parallelism for critic, you must enable `use_remove_padding`."

recipe/sppo/main_sppo.py

@@ -73,8 +73,8 @@ class TaskRunner:
         processor = hf_processor(local_path, use_fast=True)  # used for multimodal LLM, could be none
 
         # define worker classes
-        if config.actor_rollout_ref.actor.strategy == "fsdp":
-            assert config.actor_rollout_ref.actor.strategy == config.critic.strategy
+        if config.actor_rollout_ref.actor.strategy in {"fsdp", "fsdp2"}:
+            assert config.critic.strategy in {"fsdp", "fsdp2"}
             from verl.single_controller.ray import RayWorkerGroup
 
             from .sppo_worker import SPPOActorRolloutRefWorker  # , CriticWorker
@@ -115,7 +115,7 @@ class TaskRunner:
         # - finally, we combine all the rewards together
         # - The reward type depends on the tag of the data
         if config.reward_model.enable:
-            if config.reward_model.strategy == "fsdp":
+            if config.reward_model.strategy in {"fsdp", "fsdp2"}:
                 from verl.workers.fsdp_workers import RewardModelWorker
             elif config.reward_model.strategy == "megatron":
                 from verl.workers.megatron_workers import RewardModelWorker

verl/trainer/ppo/ray_trainer.py

@@ -523,7 +523,7 @@ class RayPPOTrainer:
                 assert config.critic.ppo_micro_batch_size * sp_size >= n_gpus
 
         # Check if use_remove_padding is enabled when using sequence parallelism for fsdp
-        if config.actor_rollout_ref.actor.strategy == "fsdp" and (
+        if config.actor_rollout_ref.actor.strategy in {"fsdp", "fsdp2"} and (
             config.actor_rollout_ref.actor.get("ulysses_sequence_parallel_size", 1) > 1
             or config.actor_rollout_ref.ref.get("ulysses_sequence_parallel_size", 1) > 1
         ):
@@ -531,7 +531,7 @@ class RayPPOTrainer:
                 "When using sequence parallelism for actor/ref policy, you must enable `use_remove_padding`."
             )
 
-        if self.use_critic and config.critic.strategy == "fsdp":
+        if self.use_critic and config.critic.strategy in {"fsdp", "fsdp2"}:
             if config.critic.get("ulysses_sequence_parallel_size", 1) > 1:
                 assert config.critic.model.use_remove_padding, (
                     "When using sequence parallelism for critic, you must enable `use_remove_padding`."