[trainer] fix: batch size mismatch with n>1 when gen_max for ReMax (volcengine#3779)

### 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.

Resolves volcengine#3408

We should not repeat directly on top of `gen_batch`, as in `gen_max`, we
need the original `gen_batch` so
that we can disable `do_sample` for rollout to calculate the reward
baseline.

```log
  File "verl/trainer/main_ppo.py", line 317, in run
    trainer.fit()
  File "verl/trainer/ppo/ray_trainer.py", line 1065, in fit
    batch = batch.union(gen_baseline_output)
  File "verl/protocol.py", line 802, in union
    self.batch = union_tensor_dict(self.batch, other.batch)
  File "verl/protocol.py", line 110, in union_tensor_dict
    assert tensor_dict1.batch_size == tensor_dict2.batch_size, (
AssertionError: Two tensor dict must have identical batch size. Got torch.Size([4096]) and torch.Size([16384])
```

### 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.

### 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).
(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).)

Signed-off-by: Hollow Man <[email protected]>

Author: HollowMan6

recipe/dapo/dapo_ray_trainer.py

@@ -132,14 +132,16 @@ def fit(self):
                         batch_keys=["input_ids", "attention_mask", "position_ids"],
                         non_tensor_batch_keys=["raw_prompt_ids"],
                     )
-                gen_batch = gen_batch.repeat(repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True)
+                gen_batch_output = gen_batch.repeat(
+                    repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True
+                )
 
                 is_last_step = self.global_steps >= self.total_training_steps
 
                 with marked_timer("step", timing_raw):
                     # generate a batch
                     with marked_timer("gen", timing_raw, "red"):
-                        gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch)
+                        gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch_output)
                         timing_raw.update(gen_batch_output.meta_info["timing"])
                         gen_batch_output.meta_info.pop("timing", None)
 

recipe/entropy/entropy_ray_trainer.py

@@ -108,19 +108,19 @@ def fit(self):
                         batch_keys=["input_ids", "attention_mask", "position_ids"],
                         non_tensor_batch_keys=["raw_prompt_ids"],
                     )
-                gen_batch = gen_batch.repeat(repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True)
+                gen_batch_output = gen_batch.repeat(
+                    repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True
+                )
 
                 is_last_step = self.global_steps >= self.total_training_steps
 
                 with simple_timer("step", timing_raw):
                     # generate a batch
-                    # with simple_timer("gen", timing_raw):
-                    #     gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch)
                     with simple_timer("gen", timing_raw):
                         if not self.async_rollout_mode:
-                            gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch)
+                            gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch_output)
                         else:
-                            gen_batch_output = self.async_rollout_manager.generate_sequences(gen_batch)
+                            gen_batch_output = self.async_rollout_manager.generate_sequences(gen_batch_output)
 
                     if self.config.algorithm.adv_estimator == AdvantageEstimator.REMAX:
                         with simple_timer("gen_max", timing_raw):

recipe/prime/prime_ray_trainer.py

@@ -373,12 +373,14 @@ def fit(self):
 
                 # pop those keys for generation
                 gen_batch = batch.pop(batch_keys=["input_ids", "attention_mask", "position_ids"])
-                gen_batch = gen_batch.repeat(repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True)
+                gen_batch_output = gen_batch.repeat(
+                    repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True
+                )
 
                 with simple_timer("step", timing_raw):
                     # generate a batch
                     with simple_timer("gen", timing_raw):
-                        gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch)
+                        gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch_output)
                         timing_raw.update(gen_batch_output.meta_info["timing"])
                         gen_batch_output.meta_info.pop("timing", None)
 

recipe/sppo/sppo_ray_trainer.py

@@ -182,17 +182,19 @@ def fit(self):
                     batch_keys=batch_keys_to_pop,
                     non_tensor_batch_keys=non_tensor_batch_keys_to_pop,
                 )
-                gen_batch = gen_batch.repeat(repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True)
+                gen_batch_output = gen_batch.repeat(
+                    repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True
+                )
 
                 is_last_step = self.global_steps >= self.total_training_steps
 
                 with simple_timer("step", timing_raw):
                     # generate a batch
                     with simple_timer("gen", timing_raw):
                         if not self.async_rollout_mode:
-                            gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch)
+                            gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch_output)
                         else:
-                            gen_batch_output = self.async_rollout_manager.generate_sequences(gen_batch)
+                            gen_batch_output = self.async_rollout_manager.generate_sequences(gen_batch_output)
                         timing_raw.update(gen_batch_output.meta_info["timing"])
                         gen_batch_output.meta_info.pop("timing", None)
 

verl/trainer/ppo/ray_trainer.py

@@ -1037,16 +1037,18 @@ def fit(self):
 
                 # pass global_steps to trace
                 gen_batch.meta_info["global_steps"] = self.global_steps
-                gen_batch = gen_batch.repeat(repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True)
+                gen_batch_output = gen_batch.repeat(
+                    repeat_times=self.config.actor_rollout_ref.rollout.n, interleave=True
+                )
 
                 is_last_step = self.global_steps >= self.total_training_steps
                 with marked_timer("step", timing_raw):
                     # generate a batch
                     with marked_timer("gen", timing_raw, color="red"):
                         if not self.async_rollout_mode:
-                            gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch)
+                            gen_batch_output = self.actor_rollout_wg.generate_sequences(gen_batch_output)
                         else:
-                            gen_batch_output = self.async_rollout_manager.generate_sequences(gen_batch)
+                            gen_batch_output = self.async_rollout_manager.generate_sequences(gen_batch_output)
 
                         timing_raw.update(gen_batch_output.meta_info["timing"])
                         gen_batch_output.meta_info.pop("timing", None)