[model, ci] feat: add qwen3-8b ppo script on ASCEND NPU (#3502)

### What does this PR do?

add examples/ppo_trainer/run_qwen3-8b_npu.sh

> 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

- [ ] 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.

- [ ] 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`
- [ ] 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).)

.github/workflows/e2e_ascend.yml

@@ -148,6 +148,11 @@ jobs:
           ray stop --force
           bash tests/special_npu/run_qwen3_06b_grpo.sh
           rm -rf $HOME/ckpts
+      - name: Running gsm8k e2e qwen3 training tests with PPO on ASCEND NPU
+        run: |
+          ray stop --force
+          bash tests/special_npu/run_qwen3_06b_ppo.sh
+          rm -rf $HOME/ckpts
       - name: Running gsm8k e2e training tests with GRPO MindSpeed on ASCEND NPU
         run: |
           ray stop --force

docs/ascend_tutorial/ascend_quick_start.rst

@@ -193,6 +193,8 @@ vllm & vllm-ascend
 +-----------+-------------------------+-------------+-------------------+-------------------+-------------------+--------------------------+
 |   DAPO    | Qwen3-30B-base          |    1.08%    |        pending    |        FSDP       |    vllm-ascend    |    Atlas 200T A2 Box16   |
 +-----------+-------------------------+-------------+-------------------+-------------------+-------------------+--------------------------+
+|   PPO     | Qwen3-8B                |    4.49%    |        0.874      |        FSDP       |    vllm-ascend    |    Atlas 900 A2 PODc     |
++-----------+-------------------------+-------------+-------------------+-------------------+-------------------+--------------------------+
 
 **表2** SFT类算法
 

examples/ppo_trainer/run_qwen3-8b_npu.sh

@@ -0,0 +1,55 @@
+set -x
+
+export VLLM_USE_V1=1
+
+python3 -m verl.trainer.main_ppo \
+    algorithm.adv_estimator=gae \
+    data.train_files=$HOME/data/dapo-math-17k.parquet \
+    data.val_files=$HOME/data/dapo-math-17k.parquet \
+    data.train_batch_size=256 \
+    data.max_prompt_length=2000 \
+    data.max_response_length=12000 \
+    data.shuffle=False \
+    actor_rollout_ref.model.path=Qwen/Qwen3-8B \
+    actor_rollout_ref.model.use_remove_padding=True \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.actor.ppo_mini_batch_size=64 \
+    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=1 \
+    actor_rollout_ref.actor.fsdp_config.param_offload=True \
+    actor_rollout_ref.actor.fsdp_config.optimizer_offload=True \
+    actor_rollout_ref.actor.use_kl_loss=False \
+    actor_rollout_ref.actor.ulysses_sequence_parallel_size=2 \
+    actor_rollout_ref.actor.use_dynamic_bsz=True \
+    actor_rollout_ref.actor.use_torch_compile=False \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=1 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=1 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.9 \
+    actor_rollout_ref.rollout.max_num_batched_tokens=14000 \
+    actor_rollout_ref.rollout.max_num_seqs=64 \
+    actor_rollout_ref.rollout.log_prob_use_dynamic_bsz=True \
+    actor_rollout_ref.rollout.enable_chunked_prefill=True \
+    actor_rollout_ref.rollout.enforce_eager=False \
+    critic.optim.lr=1e-5 \
+    critic.model.use_remove_padding=True \
+    critic.model.path=Qwen/Qwen3-8B \
+    critic.model.enable_gradient_checkpointing=True \
+    critic.ppo_micro_batch_size_per_gpu=1 \
+    critic.ulysses_sequence_parallel_size=2 \
+    critic.model.fsdp_config.param_offload=True \
+    critic.model.fsdp_config.optimizer_offload=True \
+    critic.use_dynamic_bsz=True \
+    trainer.critic_warmup=0 \
+    trainer.logger=console \
+    trainer.project_name='verl_example_dapo_math_17k' \
+    trainer.experiment_name='qwen3_8b_fsdp' \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=20 \
+    trainer.test_freq=-1 \
+    trainer.val_before_train=False \
+    trainer.device=npu \
+    trainer.max_actor_ckpt_to_keep=1 \
+    trainer.max_critic_ckpt_to_keep=1 \
+    trainer.total_training_steps=100 $@

tests/special_npu/run_qwen3_06b_ppo.sh

@@ -0,0 +1,51 @@
+set -x
+
+export VLLM_USE_V1=1
+
+python3 -m verl.trainer.main_ppo \
+    algorithm.adv_estimator=gae \
+    data.train_files=$HOME/data/gsm8k/train.parquet \
+    data.val_files=$HOME/data/gsm8k/test.parquet \
+    data.train_batch_size=128 \
+    data.max_prompt_length=512 \
+    data.max_response_length=128 \
+    data.shuffle=False \
+    actor_rollout_ref.model.path=Qwen/Qwen3-0.6B \
+    actor_rollout_ref.model.use_remove_padding=True \
+    actor_rollout_ref.model.enable_gradient_checkpointing=True \
+    actor_rollout_ref.actor.optim.lr=1e-6 \
+    actor_rollout_ref.actor.ppo_mini_batch_size=64 \
+    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
+    actor_rollout_ref.actor.fsdp_config.param_offload=True \
+    actor_rollout_ref.actor.fsdp_config.optimizer_offload=True \
+    actor_rollout_ref.actor.use_kl_loss=False \
+    actor_rollout_ref.actor.ulysses_sequence_parallel_size=2 \
+    actor_rollout_ref.actor.use_dynamic_bsz=True \
+    actor_rollout_ref.actor.use_torch_compile=False \
+    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=8 \
+    actor_rollout_ref.rollout.tensor_model_parallel_size=1 \
+    actor_rollout_ref.rollout.name=vllm \
+    actor_rollout_ref.rollout.gpu_memory_utilization=0.8 \
+    actor_rollout_ref.rollout.log_prob_use_dynamic_bsz=True \
+    actor_rollout_ref.rollout.enable_chunked_prefill=True \
+    actor_rollout_ref.rollout.enforce_eager=False \
+    critic.optim.lr=1e-5 \
+    critic.model.use_remove_padding=True \
+    critic.model.path=Qwen/Qwen3-0.6B \
+    critic.model.enable_gradient_checkpointing=True \
+    critic.ppo_micro_batch_size_per_gpu=8 \
+    critic.ulysses_sequence_parallel_size=2 \
+    critic.model.fsdp_config.param_offload=True \
+    critic.model.fsdp_config.optimizer_offload=True \
+    critic.use_dynamic_bsz=True \
+    trainer.critic_warmup=0 \
+    trainer.logger='["console"]' \
+    trainer.project_name='verl_ppo_example_gsm8k_qwen3' \
+    trainer.experiment_name='qwen3_06b_fsdp' \
+    trainer.n_gpus_per_node=8 \
+    trainer.nnodes=1 \
+    trainer.save_freq=-1 \
+    trainer.test_freq=5 \
+    trainer.total_epochs=1 \
+    trainer.total_training_steps=2 \
+    trainer.device=npu $@