[megatron] feat: support discrete profiling for mindspeed (#4271)

### What does this PR do?

1. add role in annotate call  for megatron_workers.py
2. update doc

#4206

> 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: ...
- [ ] 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.

### 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).
- [ ] 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).
- [ ] 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).)

---------

Co-authored-by: mengchengTang <[email protected]>
Co-authored-by: Shangwei-Li <[email protected]>

Author: 3 people

docs/ascend_tutorial/ascend_profiling_en.rst

@@ -1,10 +1,10 @@
-Data collection based on FSDP backend on Ascend devices(en)
+Performance data collection based on FSDP or MindSpeed(Megatron) on Ascend devices(en)
 ==========================================================================================
 
 Last updated: 08/14/2025.
 
 This is a tutorial for data collection using the GRPO or DAPO algorithm
-based on FSDP on Ascend devices.
+based on FSDP or MindSpeed(Megatron) on Ascend devices.
 
 Configuration
 -------------

docs/ascend_tutorial/ascend_profiling_zh.rst

@@ -1,11 +1,11 @@
-Data collection based on FSDP backend on Ascend devices(zh)
+Performance data collection based on FSDP or MindSpeed(Megatron) on Ascend devices(zh)
 ====================================
 
-在昇腾设备上基于FSDP后端进行数据采集
+在昇腾设备上基于FSDP或MindSpeed(Megatron)后端进行性能数据采集
 
 Last updated: 08/14/2025.
 
-这是一份在昇腾设备上基于FSDP后端使用GRPO或DAPO算法进行数据采集的教程。
+这是一份在昇腾设备上基于FSDP或MindSpeed(Megatron)后端，使用GRPO或DAPO算法进行数据采集的教程。
 
 配置
 ----

verl/workers/fsdp_workers.py

@@ -1486,7 +1486,7 @@ def init_model(self):
         )
 
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="critic"))
-    @DistProfiler.annotate(color="cyan")
+    @DistProfiler.annotate(color="cyan", role="compute_values")
     def compute_values(self, data: DataProto):
         if self._is_offload_param:
             load_fsdp_model_to_gpu(self.critic_module)
@@ -1506,7 +1506,7 @@ def compute_values(self, data: DataProto):
         return output
 
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="critic"))
-    @DistProfiler.annotate(color="pink")
+    @DistProfiler.annotate(color="pink", role="critic_update")
     def update_critic(self, data: DataProto):
         if self._is_offload_param:
             load_fsdp_model_to_gpu(self.critic_module)
@@ -1874,7 +1874,7 @@ def _switch_chat_template(self, data: DataProto):
         return DataProto.from_dict(rm_inputs)
 
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="reward"))
-    @DistProfiler.annotate(color="brown")
+    @DistProfiler.annotate(color="brown", role="compute_rm_score")
     def compute_rm_score(self, data: DataProto):
         import itertools
 

verl/workers/megatron_workers.py

@@ -724,7 +724,7 @@ async def trainer_mode(self):
 
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="actor"))
     @GPUMemoryLogger(role="update_actor", logger=logger)
-    @DistProfiler.annotate(color="red")
+    @DistProfiler.annotate(color="red", role="actor_update")
     def update_actor(self, data: DataProto):
         assert self._is_actor
         if self._is_offload_param:
@@ -767,7 +767,7 @@ def update_actor(self, data: DataProto):
 
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="rollout"))
     @GPUMemoryLogger(role="generate_sequences", logger=logger)
-    @DistProfiler.annotate(color="red")
+    @DistProfiler.annotate(color="red", role="rollout_generate")
     def generate_sequences(self, prompts: DataProto):
         assert self._is_rollout
         prompts = prompts.to(get_device_name())
@@ -817,7 +817,7 @@ def generate_sequences(self, prompts: DataProto):
 
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="actor"))
     @GPUMemoryLogger(role="compute_ref_log_prob", logger=logger)
-    @DistProfiler.annotate(color="olive")
+    @DistProfiler.annotate(color="olive", role="ref_compute_log_prob")
     def compute_ref_log_prob(self, data: DataProto):
         assert self._is_ref
         if self._ref_is_offload_param:
@@ -839,7 +839,7 @@ def compute_ref_log_prob(self, data: DataProto):
 
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="actor"))
     @GPUMemoryLogger(role="compute_log_prob", logger=logger)
-    @DistProfiler.annotate(color="blue")
+    @DistProfiler.annotate(color="blue", role="actor_compute_log_prob")
     def compute_log_prob(self, data: DataProto):
         assert self._is_actor
         if self._is_offload_param:
@@ -1207,7 +1207,7 @@ def init_model(self):
         )
 
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="critic"))
-    @DistProfiler.annotate(color="cyan")
+    @DistProfiler.annotate(color="cyan", role="compute_values")
     def compute_values(self, data: DataProto):
         micro_batch_size = self.config.ppo_micro_batch_size_per_gpu
         data.meta_info["micro_batch_size"] = micro_batch_size
@@ -1224,7 +1224,7 @@ def compute_values(self, data: DataProto):
         return output
 
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="critic"))
-    @DistProfiler.annotate(color="pink")
+    @DistProfiler.annotate(color="pink", role="critic_update")
     def update_critic(self, data: DataProto):
         data = data.to(get_device_id())
 
@@ -1448,7 +1448,7 @@ def init_model(self):
     # TODO: reward model use itself tokenizer instead of sft tokenizer
     # the input_ids, responses, attention_mask and position_ids may be different!
     @register(dispatch_mode=make_nd_compute_dataproto_dispatch_fn(mesh_name="reward"))
-    @DistProfiler.annotate(color="brown")
+    @DistProfiler.annotate(color="brown", role="compute_rm_score")
     def compute_rm_score(self, data: DataProto):
         data.meta_info["micro_batch_size"] = self.config.micro_batch_size_per_gpu
         data.meta_info["max_token_len"] = self.config.forward_max_token_len_per_gpu