[veomni] refactor: no long check the attn_implementation/moe_implementation in VeOmniEngineConfig (#5019)

in case VeOmni backend add more attn_implementation and
moe_implementation, no long check the
attn_implementation/moe_implementation in VeOmniEngineConfig, and
leaving the work for `build_foundation_model`, which is interface VeOmni
provided.

### 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`, `veomni`, `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).)
- [ ] If your PR is related to the `recipe` submodule, please also
update the reference to the submodule commit via `git submodule update
--remote` or `cd recipe && git pull origin main`.

---------

Co-authored-by: Huazhong <hzji210@gmail.com>

Author: A1waysBeenHere

verl/trainer/config/engine/veomni.yaml

@@ -56,9 +56,12 @@ enable_fsdp_offload: false
 
 enable_reentrant: false
 
+# support eager, sdpa, flash_attention_2, flash_attention_3, veomni_flash_attention_2_with_sp,
+# veomni_flash_attention_3_with_sp and native-sparse
 attn_implementation: flash_attention_2
 
-moe_implementation: eager
+# eager or fused
+moe_implementation: fused
 
 force_use_huggingface: false
 

verl/workers/config/engine.py

@@ -212,8 +212,21 @@ class VeOmniEngineConfig(EngineConfig):
         enable_full_shard (bool): Enable fully shard for FSDP training (ZeRO-3), default False
         enable_fsdp_offload (bool): Enable CPU offload for FSDP1, default False
         enable_reentrant (bool): Use reentrant gradient checkpointing, default False
-        attn_implementation (str): Attention implementation to use, default "flash-attn"
-        moe_implementation (str): MoE implementation to use, default "eager"
+        attn_implementation (str): Attention implementation to use.
+            1. `eager`
+            2. `sdpa`
+            3. `flash_attention_2`
+            4. `flash_attention_3`
+            5. `veomni_flash_attention_2_with_sp`
+            6. `veomni_flash_attention_3_with_sp`
+            7. `native-sparse`
+            default "flash_attention_2"
+            Note: In case VeOmni add more attn_implementation, please check https://github.com/ByteDance-Seed/VeOmni/
+        moe_implementation (str): MoE implementation to use.
+            1. `eager`
+            2. `fused`
+            default "fused"
+            Note: In case VeOmni add more moe_implementation, please check https://github.com/ByteDance-Seed/VeOmni/
         force_use_huggingface (bool): Force loading model from huggingface, default False
         activation_gpu_limit (float): When enabling activation offload, `activation_gpu_limit` GB
             activations are allowed to reserve on GPU, default 0.0
@@ -256,8 +269,8 @@ class VeOmniEngineConfig(EngineConfig):
     load_checkpoint_path: Optional[str] = None
     enable_fsdp_offload: bool = False
     enable_reentrant: bool = False
-    attn_implementation: Optional[Literal["eager", "sdpa", "flash_attention_2", "native-sparse"]] = "flash_attention_2"
-    moe_implementation: Optional[Literal[None, "eager", "fused"]] = "eager"
+    attn_implementation: str = "flash_attention_2"
+    moe_implementation: str = "fused"
     force_use_huggingface: bool = False
     activation_gpu_limit: float = 0.0
     basic_modules: Optional[list[str]] = field(default_factory=list)