frozenleaves/verl
1
0
Fork 0
mirror of https://github.com/volcengine/verl.git synced 2025-10-20 13:43:50 +08:00
Code Issues Packages Projects Releases Wiki Activity
1,505 Commits 15 Branches 12 Tags
main
Commit Graph

365 Commits

Author SHA1 Message Date
Yingru Li
4f1c489e45 [algo] fix: remove torch.quantile-based percentile metrics to resolve tensor size limit error (#3810) 
## Summary

Fixes #3787 by removing `torch.quantile()`-based percentile metrics
(`rollout_is_p25`, `rollout_is_p50`, `rollout_is_p75`) that caused
`RuntimeError: quantile() input tensor is too large` when using large
batch sizes or response lengths.

## Problem

When using configurations with large tensor sizes (e.g.,
`max_response_length: 32k`, `rollout.n: 16`, `train_batch_size: 16`),
the `torch.quantile()` function fails with a runtime error due to
PyTorch's internal tensor size limitations (~2^24 to 2^27 elements
depending on version, GPU memory, and dtype).

The error occurred in `verl/trainer/ppo/mismatch_helper.py`:
```python
metrics["rollout_is_p25"] = torch.quantile(flat_weights, 0.25)
metrics["rollout_is_p50"] = torch.quantile(flat_weights, 0.50)
metrics["rollout_is_p75"] = torch.quantile(flat_weights, 0.75)
```

## Solution

Removed the three quantile-based percentile metrics from the Rollout IS
framework. The remaining metrics (`rollout_is_mean`, `rollout_is_std`,
`rollout_is_min`, `rollout_is_max`, `rollout_is_eff_sample_size`, etc.)
provide sufficient monitoring capabilities for importance sampling
health without triggering tensor size limitations.

## Changes

- **Modified**:
[verl/trainer/ppo/mismatch_helper.py](verl/trainer/ppo/mismatch_helper.py)
- Removed `rollout_is_p25`, `rollout_is_p50`, `rollout_is_p75` metric
calculations
  - All other rollout IS and mismatch metrics remain functional

## Testing

Verified that:
- Rollout IS framework continues to function correctly without
percentile metrics
- No runtime errors with large tensor configurations
- All other metrics (mean, std, min, max, ESS, veto fraction, etc.) are
computed correctly

Resolves #3787
 2025-10-20 13:04:57 +08:00
OC
53aed3eea1 [doc] fix: update install instruction and retool readme (#3824) 
### What does this PR do?

Update docs according to user feedback:
1. add a note: dockerfiles provide more detailed installation steps than
installation doc
2. add a "how it works" section in retool readme to avoid training
without SFT.

### 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).)
 2025-10-20 11:43:11 +08:00
ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟
65eb019a81 [trainer] fix: Add data.seed to config (#3815) 2025-10-20 09:57:14 +08:00
arron
85d5b2ee2e [doc] feat: update fully async experiment message (#3804) 2025-10-18 06:20:01 +08:00
arron
b25bb7d4f3 [trainer, recipe] feat: fully async training recipe (#2981) 
### What does this PR do?

To implement a purely asynchronous training workflow, we further split
the training process into a Trainer and a Rollouter based on the
existing one-step-off policy code, with samples transmitted via a
message queue.

We will continue to integrate partial rollout to mitigate the impact of
long-tail training.

> Add **concise** overview of what this PR aims to achieve or
accomplish. Reference related GitHub issues and PRs that help with the
review. https://github.com/volcengine/verl/pull/2231
https://github.com/volcengine/verl/pull/2200

### 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).)

---------

Co-authored-by: meituan-search <machi04@meituan.com>
Co-authored-by: wangshulin02 <wangshulin02@meituan.com>
Co-authored-by: arron <arron@MBP-2G17FXQ05P-2332.local>
Co-authored-by: wangshulin02 <953550366@qq.com>
Co-authored-by: hadoop-ai-search <hadoop-ai-search@set-zw04-mlp-codelab-pc1189.mt>
Co-authored-by: sl-1314 <82856253+sl-1314@users.noreply.github.com>
Co-authored-by: arron <arron@MBP-VH9RV7LTJC-1907.local>
Co-authored-by: arron <arron@MBP-JFQXPWR11F-1943.local>
 2025-10-17 22:29:18 +08:00
ℍ𝕠𝕝𝕝𝕠𝕨 𝕄𝕒𝕟
acfcf98ed0 [doc] fix: actor_rollout_ref.critic is not correct (#3778) 
### 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.

They should start directly with `critic`

### 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 <hollowman@opensuse.org>
 2025-10-16 11:12:45 +08:00
Yan Bai
33eb86f54f [megatron] feat: support qwen3vl (#3763) 
### 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.

support training qwen3vl with megatron

1. add an image with vllm0.11 and nemo's dedicated megatron that support
gpt-oss with optimized fused kernels.
2. add a script of training qwen3vl-30b with megatron
3. necessary changes to support qwen3vl megatron. (just register forward
functions, the modeling is through mbridge)


### 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.
<img width="372" height="314" alt="image"
src="https://github.com/user-attachments/assets/f1126e46-51a9-4e00-958f-5d034b8f94bd"
/>

### 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).)
 2025-10-15 10:19:22 +08:00
Yingru Li
5d378b5f95 [rollout] refactor: rename "clip" mode back to "mask" mode (#3750) 
# Rollout Importance Sampling Framework

related to https://github.com/volcengine/verl/pull/3694

## Summary

This PR introduces a comprehensive **Rollout Importance Sampling (IS)**
framework to correct distribution mismatch between data-collecting
(rollout) and training policies, a critical factor for ensuring stable
and efficient model training in RL fine-tuning.

This work is motivated by the analysis in our blog post, [When Speed
Kills Stability: Demystifying RL Collapse from the Inference-Training
Mismatch](https://yingru.notion.site/When-Speed-Kills-Stability-271211a558b7808d8b12d403fd15edda).
If you find this implementation useful in your research, please consider
citing:

```bibtex
@misc{liu-li-2025,
  title = {When Speed Kills Stability: Demystifying RL Collapse from the Inference-Training Mismatch},
  url = {https://yingru.notion.site/When-Speed-Kills-Stability-Demystifying-RL-Collapse-from-the-Inference-Training-Mismatch-271211a558b7808d8b12d403fd15edda},
  author = {Jiacai Liu and Yingru Li and Yuqian Fu and Jiawei Wang and Qian Liu and Yu Shen},
  year = {2025},
  month = {September},
}
```

---

## Problem Statement

When using different policies for rollout generation (e.g., vLLM with
BFloat16) and training (e.g., FSDP with FP32), distribution mismatch
occurs, leading to:
- Biased gradient estimates
- Training instability and collapse
- Reduced sample efficiency
- Poor convergence properties

This framework addresses these issues through principled importance
sampling correction.

---

## Key Features & Improvements

### 1. **Flexible Aggregation Levels**
Three methods for calculating IS weights:
- **`token`**: Per-token importance ratios
- **`sequence`**: Product of per-token ratios
- **`geometric`**: Geometric mean of ratios

### 2. **Advanced Bounding Modes**
Two strategies to control weight variance:
- **`truncate`** (TIS): Caps weights at upper threshold only, preserving
gradients
- **`mask`** (MIS): Zeros out weights outside bounds, more aggressive
filtering

### 3. **Comprehensive Diagnostics**
Detailed metrics to monitor distribution mismatch and training health:

**Rollout IS Metrics** (automatically prefixed with `mismatch/`):
- Health indicators: `rollout_is_eff_sample_size`, `rollout_is_mean`
- Distribution statistics: `rollout_is_p25`, `rollout_is_p50`,
`rollout_is_p75`, `rollout_is_p95`, `rollout_is_p99`, `rollout_is_max`,
`rollout_is_min`, `rollout_is_std`
- Diagnostics: `rollout_is_veto_fraction`,
`rollout_is_catastrophic_token_fraction`, `rollout_is_masked_fraction`
(mask mode)
- Sequence-level statistics (for sequence/geometric modes):
`rollout_is_seq_mean`, `rollout_is_seq_std`, `rollout_is_seq_max`,
`rollout_is_seq_min`, etc.

**Mismatch Metrics** (computed efficiently within IS weight
computation):
- KL Divergence: `mismatch_kl` (forward KL), `mismatch_k3_kl` (K3
estimator for stability)
- Perplexity: `mismatch_training_ppl`, `mismatch_rollout_ppl`,
`mismatch_ppl_ratio`
- Log perplexity statistics: `mismatch_log_ppl_diff`,
`mismatch_log_ppl_abs_diff`, `mismatch_log_ppl_diff_max`,
`mismatch_log_ppl_diff_min`

### 4. **Outlier Mitigation**
- **Veto mechanism**: Automatically discards samples with catastrophic
importance weights (per-token ratios below threshold)
- Prevents gradient corruption from extreme outliers
- Configurable threshold (default: 1e-4)

### 5. **Numerical Stability**
- All core computations in **log-space** to prevent underflow/overflow
- Carefully designed clamping and bounding to maintain numerical
precision
- Safe handling of edge cases (zero probabilities, extreme ratios)

### 6. **Memory Efficiency**
- Optimized computation to minimize CUDA memory usage
- Efficient metric aggregation without large intermediate tensors
- Suitable for large-scale distributed training

### 7. **Metrics-Only Mode**
- Compute and monitor mismatch metrics **without** applying IS weights
- Useful for:
  - Understanding distribution mismatch before intervention
  - Deciding whether IS correction is needed
  - A/B testing IS impact
- Controlled by `algorithm.rollout_is` flag (independent of weight
computation)

### 8. **Universal PPO Support**
- Integrated with **all PPO variants**: vanilla, GSPO, GPG, Clip-Cov,
KL-Cov, geo_mean
- Consistent interface across different policy loss functions
- Automatic weight application when enabled

---

## API and Configuration Changes

### Migration from Legacy TIS

#### ❌ **Before (REMOVED)**
```yaml
# Old TIS configuration - NO LONGER SUPPORTED
actor_rollout_ref:
  actor:
    tis_imp_ratio_cap: 2.0  # Removed from actor config
```

The legacy implementation:
- Only supported token-level truncation
- No metrics tracking
- Lacked numerical stability
- Limited configurability

#### ✅ **After (New Framework)**

Configuration moved to `algorithm` section for better organization:

```yaml
algorithm:
  # Main on/off switch: null = disabled, float = enabled
  rollout_is_threshold: 2.0

  # Control weight application (independent of metrics computation)
  rollout_is: true  # true = apply weights, false = metrics only

  # Optional: lower threshold (defaults to 1/upper if null)
  rollout_is_threshold_lower: null

  # Aggregation level: "token", "sequence", or "geometric"
  rollout_is_level: token

  # Bounding mode: "truncate" or "mask"
  rollout_is_mode: truncate

  # Veto threshold for catastrophic outliers (null = disabled)
  rollout_is_veto_threshold: 1e-4

# REQUIRED: Enable log probability calculation
actor_rollout_ref:
  rollout:
    calculate_log_probs: true
```

### Configuration Examples

**1. Token-level truncation (recommended starting point)**
```yaml
algorithm:
  rollout_is_threshold: 2.0
  rollout_is: true
  rollout_is_level: token
  rollout_is_mode: truncate
```

**2. Sequence-level masking (more aggressive)**
```yaml
algorithm:
  rollout_is_threshold: 2.0
  rollout_is: true
  rollout_is_level: sequence
  rollout_is_mode: mask
```

**3. Metrics-only mode (monitoring without correction)**
```yaml
algorithm:
  rollout_is_threshold: 2.0
  rollout_is: false  # Compute metrics but don't apply weights
  rollout_is_level: token
  rollout_is_mode: truncate
```

**Example script:** `bash
examples/rollout_importance_sampling/run_with_rollout_is.sh`

---

## Code Changes Overview

### New Files (4 files, 1,442 lines)

1. **`verl/trainer/ppo/mismatch_helper.py`** (459 lines)
   - Core implementation of IS weight computation
   - Three aggregation levels: token, sequence, geometric
   - Two bounding modes: truncate, mask
   - Veto mechanism for outlier detection
   - Comprehensive metrics computation (IS + mismatch)
   - All computations in log-space for numerical stability
   - Memory-efficient design

2. **`docs/advance/rollout_is_migration.md`** (642 lines)
   - Comprehensive migration guide from legacy TIS
   - Detailed explanation of all configuration options
   - Recommended threshold ranges for each aggregation level
   - Troubleshooting guide and best practices
   - Metrics interpretation guide

3. **`examples/rollout_importance_sampling/README.md`** (242 lines)
   - Quick start guide with working examples
   - Configuration templates for common scenarios
   - Threshold tuning guidelines
   - Metrics monitoring instructions

4. **`examples/rollout_importance_sampling/run_with_rollout_is.sh`** (99
lines)
   - Complete working example script
   - Demonstrates token-level and sequence-level configurations
   - Ready to run with minimal modifications

### Modified Core Files (9 files)

1. **`verl/trainer/ppo/core_algos.py`** (~50 lines changed)
   - Removed legacy TIS logic (`tis_imp_ratio_cap`)
   - Added `rollout_is_weights` parameter to all policy loss functions
   - Unified IS weight application interface across all PPO variants:
     - `compute_policy_loss_vanilla`
     - `compute_policy_loss_gspo`
     - `compute_policy_loss_gpg`
     - `compute_policy_loss_clip_cov`
     - `compute_policy_loss_kl_cov`
     - `compute_policy_loss_geo_mean`
   - Special handling for `geo_mean` (sequence-level aggregation)

2. **`verl/trainer/ppo/ray_trainer.py`** (~52 lines added)
   - New method: `compute_rollout_importance_weights_and_add_to_batch()`
   - Centralized IS computation (once per batch, on driver)
- Conditional weight distribution to workers based on
`algorithm.rollout_is`
   - Metrics collection and aggregation
   - Integration with existing training loop

3. **`verl/trainer/config/algorithm.py`** (+18 lines)
   - Added 6 new Rollout IS parameters:
     - `rollout_is_threshold` (main on/off switch)
     - `rollout_is` (weight application control)
     - `rollout_is_threshold_lower`
     - `rollout_is_level`
     - `rollout_is_mode`
     - `rollout_is_veto_threshold`
   - Comprehensive docstrings explaining each parameter

4. **`verl/workers/config/actor.py`** (-1 line)
   - Removed deprecated `tis_imp_ratio_cap` parameter

5. **`verl/workers/actor/dp_actor.py`** (~26 lines changed)
   - Updated to use new `rollout_is_weights` parameter
   - Removed legacy TIS logic

6. **`verl/workers/actor/megatron_actor.py`** (~15 lines changed)
   - Updated to use new `rollout_is_weights` parameter
   - Removed legacy TIS logic

7. **Configuration Files** (4 files updated)
   - `verl/trainer/config/ppo_trainer.yaml`
   - `verl/trainer/config/ppo_megatron_trainer.yaml`
   - `verl/trainer/config/_generated_ppo_trainer.yaml`
   - `verl/trainer/config/_generated_ppo_megatron_trainer.yaml`
- Added default Rollout IS configuration section with explanatory
comments

### Testing (2 files, 530 lines)

1. **`tests/trainer/ppo/test_rollout_is.py`** (289 lines)
   - Unit tests for `mismatch_helper.py`
   - Coverage for all aggregation levels (token, sequence, geometric)
   - Coverage for all bounding modes (truncate, mask)
   - Veto mechanism tests
   - Edge case handling (zeros, extremes, empty sequences)
   - Numerical stability verification
   - Metrics correctness validation

2. **`tests/trainer/ppo/test_rollout_is_integration.py`** (241 lines)
   - Integration tests with PPO training loop
   - End-to-end workflow validation
   - Batch processing tests
   - Configuration validation
   - Metrics collection verification
   - Compatibility with distributed training

### Updated Recipes (2 files)

1. **`recipe/dapo/dapo_ray_trainer.py`** (+5 lines)
   - Updated imports to use new framework

2. **`recipe/dapo/run_dapo_qwen2.5_32b_tis.sh`** (~42 lines changed)
   - Migrated from legacy TIS to new Rollout IS configuration
   - Updated documentation and comments

### Documentation Updates (2 files)

1. **`docs/examples/config.rst`** (~22 lines changed)
   - Updated configuration examples
   - Added Rollout IS section

2. **`docs/index.rst`** (+1 line)
   - Added link to Rollout IS migration guide

---

## Implementation Highlights

### Centralized Architecture

The new design follows a clean separation of concerns:

```
ray_trainer.py (driver)
    └─> compute_rollout_importance_weights_and_add_to_batch()
         └─> mismatch_helper.compute_rollout_importance_weights()
              ├─> Computes IS weights (token/sequence/geometric)
              ├─> Applies bounding (truncate/mask)
              ├─> Veto mechanism for outliers
              ├─> Computes IS metrics
              └─> Computes mismatch metrics (KL, PPL)
    └─> Conditionally adds weights to batch (if rollout_is=True)
    └─> Distributes batch to workers

actor workers (dp_actor, megatron_actor)
    └─> Receive batch with rollout_is_weights (if enabled)
    └─> Pass weights to policy loss function

core_algos.py
    └─> All policy loss functions accept rollout_is_weights
    └─> Apply weights if provided: pg_losses *= rollout_is_weights
```

### Key Design Decisions

1. **Centralized Computation**: IS weights computed once on driver, not
per worker
   - Reduces redundant computation
   - Ensures consistency across workers
   - Simplifies debugging and metrics collection

2. **Configuration in Algorithm**: Moved from actor config to algorithm
config
- Better conceptual organization (algorithm-level concern, not
worker-level)
   - Easier to manage and validate
   - Consistent with other algorithm parameters

3. **Two-Level Control**:
   - `rollout_is_threshold`: Enables/disables entire system (null = off)
- `rollout_is`: Controls weight application (true = apply, false =
metrics only)
   - Allows flexible monitoring and gradual rollout

4. **Metrics Consolidation**: Mismatch metrics computed within IS weight
computation
   - Eliminates duplicate computation
   - Reduces memory overhead
   - Maintains metric accuracy

5. **Universal PPO Support**: Single interface for all PPO variants
   - Minimal code changes required
   - Consistent behavior across algorithms
   - Easy to add new variants

---

## Migration Guide

### For Users of Legacy TIS

**Step 1: Update your configuration file**

```yaml
# OLD (remove this)
actor_rollout_ref:
  actor:
    tis_imp_ratio_cap: 2.0

# NEW (add this)
algorithm:
  rollout_is_threshold: 2.0  # Use same value as old tis_imp_ratio_cap
  rollout_is: true
  rollout_is_level: token
  rollout_is_mode: truncate

# REQUIRED (add if not present)
actor_rollout_ref:
  rollout:
    calculate_log_probs: true
```

**Step 2: Monitor metrics**

The first time you run with the new configuration, check these metrics:
- `mismatch/rollout_is_eff_sample_size`: Should be > 80% of batch size
- `mismatch/rollout_is_veto_fraction`: Should be < 5%
- `mismatch/rollout_is_mean`: Should be close to 1.0

**Step 3: Tune if needed**

If effective sample size is too low:
- Increase `rollout_is_threshold`
- Try `rollout_is_mode: mask` with appropriate lower bound
- Consider `rollout_is_level: sequence` for more aggressive correction

For detailed guidance, see `docs/advance/rollout_is_migration.md`.

### For New Users

Start with recommended defaults:

```yaml
algorithm:
  rollout_is_threshold: 2.0
  rollout_is: true
  rollout_is_level: token
  rollout_is_mode: truncate

actor_rollout_ref:
  rollout:
    calculate_log_probs: true
```

Run the example script to see it in action:
```bash
bash examples/rollout_importance_sampling/run_with_rollout_is.sh
```

---

## Testing

### Unit Tests
- **289 lines** of comprehensive unit tests in `test_rollout_is.py`
- Covers all aggregation levels, bounding modes, and edge cases
- Validates numerical stability and correctness
- Fast execution (~1-2 seconds)

### Integration Tests
- **241 lines** of integration tests in `test_rollout_is_integration.py`
- End-to-end workflow with PPO training loop
- Distributed training compatibility
- Metrics collection validation
- Moderate execution time (~10-20 seconds)

### Running Tests
```bash
# Run all Rollout IS tests
pytest tests/trainer/ppo/test_rollout_is.py -v
pytest tests/trainer/ppo/test_rollout_is_integration.py -v

# Run specific test
pytest tests/trainer/ppo/test_rollout_is.py::test_token_level_truncate -v
```

---

## Metrics Reference

### Rollout IS Metrics (all prefixed with `mismatch/`)

| Metric | Description | Ideal Range |
|--------|-------------|-------------|
| `rollout_is_eff_sample_size` | Effective number of samples after IS |
> 80% of batch |
| `rollout_is_mean` | Mean IS weight | ~1.0 |
| `rollout_is_std` | Standard deviation of IS weights | Low variance |
| `rollout_is_p25` | 25th percentile | ~0.8-1.0 |
| `rollout_is_p50` | Median IS weight | ~1.0 |
| `rollout_is_p75` | 75th percentile | ~1.0-1.2 |
| `rollout_is_p95` | 95th percentile | < threshold |
| `rollout_is_p99` | 99th percentile | < threshold |
| `rollout_is_max` | Maximum weight | ≤ threshold |
| `rollout_is_min` | Minimum weight | ≥ lower threshold (mask mode) |
| `rollout_is_veto_fraction` | % sequences vetoed | < 5% |
| `rollout_is_catastrophic_token_fraction` | % catastrophic tokens | <
1% |
| `rollout_is_masked_fraction` | % tokens masked (mask mode) | Variable
|

### Mismatch Metrics (all prefixed with `mismatch/`)

| Metric | Description | What It Means |
|--------|-------------|---------------|
| `mismatch_kl` | Forward KL divergence | Distribution difference
(rollout vs training) |
| `mismatch_k3_kl` | K3 KL estimator | Stable KL estimate for small
divergences |
| `mismatch_training_ppl` | Training policy perplexity | Prediction
difficulty of training policy |
| `mismatch_rollout_ppl` | Rollout policy perplexity | Prediction
difficulty of rollout policy |
| `mismatch_ppl_ratio` | Ratio of training to rollout PPL | Relative
prediction difficulty |
| `mismatch_log_ppl_diff` | Log perplexity difference | Sequence-level
PPL mismatch |
| `mismatch_log_ppl_abs_diff` | Absolute log PPL difference | Magnitude
of mismatch |
| `mismatch_log_ppl_diff_max` | Max log PPL difference | Worst-case
mismatch |
| `mismatch_log_ppl_diff_min` | Min log PPL difference | Best-case
mismatch |
| `mismatch_training_log_ppl` | Log of training PPL | Log-scale training
perplexity |
| `mismatch_rollout_log_ppl` | Log of rollout PPL | Log-scale rollout
perplexity |

---

## Performance Impact

### Memory
- Minimal overhead: ~1-2% increase in peak memory usage
- Efficient log-space computation
- No large intermediate tensors

### Computation
- Negligible impact on training speed: < 1% overhead
- Centralized computation on driver (no per-worker redundancy)
- Optimized tensor operations

### Training Stability
- Significant improvement in stability when distribution mismatch exists
- Faster convergence in many scenarios
- Reduced risk of training collapse

---

## Breaking Changes

> [!IMPORTANT]
> This PR contains **BREAKING CHANGES** to the configuration API.

### Removed
- `actor_rollout_ref.actor.tis_imp_ratio_cap`: No longer supported

### Migration Required
All users of the legacy TIS implementation must update their
configuration files. See the migration guide above or
`docs/advance/rollout_is_migration.md` for detailed instructions.

### Backward Compatibility
- No backward compatibility with legacy TIS
- Configuration files with `tis_imp_ratio_cap` will raise validation
errors
- Affected recipes have been updated in this PR

---

## Pre-Submission Checklist

- [x] Search for similar PRs:
[https://github.com/volcengine/verl/pulls?q=is%3Apr+importance+sampling](https://github.com/volcengine/verl/pulls?q=is%3Apr+importance+sampling)
- [x] Format PR title as `[{modules}] {type}: {description}` (checked by
CI)
- **Suggested title:** `[BREAKING][rollout, trainer, algo] feat:
implement comprehensive Rollout Importance Sampling framework`
- [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)
- [x] Add/update
[documentation](https://github.com/volcengine/verl/tree/main/docs) (3
new docs, 2 updated)
- [x] Add unit and integration tests (530 lines of tests)
- [x] Once PR is ready for CI, send message in `ci-request` channel

---

## References

- **Blog post:** [When Speed Kills Stability: Demystifying RL Collapse
from the Inference-Training
Mismatch](https://yingru.notion.site/When-Speed-Kills-Stability-271211a558b7808d8b12d403fd15edda)
- **Migration guide:** `docs/advance/rollout_is_migration.md`
- **Examples:** `examples/rollout_importance_sampling/`
- **Tests:** `tests/trainer/ppo/test_rollout_is*.py`
 2025-10-13 11:06:36 -07:00
Yingru Li
21271aabb9 [BREAKING][rollout, trainer, algo] feat: comprehensive rollout importance sampling implementation (#3694) 
# Rollout Importance Sampling Framework

## Summary

This PR introduces a comprehensive **Rollout Importance Sampling (IS)**
framework to correct distribution mismatch between data-collecting
(rollout) and training policies, a critical factor for ensuring stable
and efficient model training in RL fine-tuning.

This work is motivated by the analysis in our blog post, [When Speed
Kills Stability: Demystifying RL Collapse from the Inference-Training
Mismatch](https://yingru.notion.site/When-Speed-Kills-Stability-271211a558b7808d8b12d403fd15edda).
If you find this implementation useful in your research, please consider
citing:

```bibtex
@misc{liu-li-2025,
  title = {When Speed Kills Stability: Demystifying RL Collapse from the Inference-Training Mismatch},
  url = {https://yingru.notion.site/When-Speed-Kills-Stability-Demystifying-RL-Collapse-from-the-Inference-Training-Mismatch-271211a558b7808d8b12d403fd15edda},
  author = {Jiacai Liu and Yingru Li and Yuqian Fu and Jiawei Wang and Qian Liu and Yu Shen},
  year = {2025},
  month = {September},
}
```

---

## Problem Statement

When using different policies for rollout generation (e.g., vLLM with
BFloat16) and training (e.g., FSDP with FP32), distribution mismatch
occurs, leading to:
- Biased gradient estimates
- Training instability and collapse
- Reduced sample efficiency
- Poor convergence properties

This framework addresses these issues through principled importance
sampling correction.

---

## Key Features & Improvements

### 1. **Flexible Aggregation Levels**
Three methods for calculating IS weights:
- **`token`**: Per-token importance ratios
- **`sequence`**: Product of per-token ratios
- **`geometric`**: Geometric mean of ratios

### 2. **Advanced Bounding Modes**
Two strategies to control weight variance:
- **`truncate`** (TIS): Caps weights at upper threshold only, preserving
gradients
- **`clip`** (CIS): Zeros out weights outside bounds, more aggressive
filtering

### 3. **Comprehensive Diagnostics**
Detailed metrics to monitor distribution mismatch and training health:

**Rollout IS Metrics** (automatically prefixed with `mismatch/`):
- Health indicators: `rollout_is_eff_sample_size`, `rollout_is_mean`
- Distribution statistics: `rollout_is_p25`, `rollout_is_p50`,
`rollout_is_p75`, `rollout_is_p95`, `rollout_is_p99`, `rollout_is_max`,
`rollout_is_min`, `rollout_is_std`
- Diagnostics: `rollout_is_veto_fraction`,
`rollout_is_catastrophic_token_fraction`, `rollout_is_clipped_fraction`
(clip mode)
- Sequence-level statistics (for sequence/geometric modes):
`rollout_is_seq_mean`, `rollout_is_seq_std`, `rollout_is_seq_max`,
`rollout_is_seq_min`, etc.

**Mismatch Metrics** (computed efficiently within IS weight
computation):
- KL Divergence: `mismatch_kl` (forward KL), `mismatch_k3_kl` (K3
estimator for stability)
- Perplexity: `mismatch_training_ppl`, `mismatch_rollout_ppl`,
`mismatch_ppl_ratio`
- Log perplexity statistics: `mismatch_log_ppl_diff`,
`mismatch_log_ppl_abs_diff`, `mismatch_log_ppl_diff_max`,
`mismatch_log_ppl_diff_min`

### 4. **Outlier Mitigation**
- **Veto mechanism**: Automatically discards samples with catastrophic
importance weights (per-token ratios below threshold)
- Prevents gradient corruption from extreme outliers
- Configurable threshold (default: 1e-4)

### 5. **Numerical Stability**
- All core computations in **log-space** to prevent underflow/overflow
- Carefully designed clipping and bounding to maintain numerical
precision
- Safe handling of edge cases (zero probabilities, extreme ratios)

### 6. **Memory Efficiency**
- Optimized computation to minimize CUDA memory usage
- Efficient metric aggregation without large intermediate tensors
- Suitable for large-scale distributed training

### 7. **Metrics-Only Mode**
- Compute and monitor mismatch metrics **without** applying IS weights
- Useful for:
  - Understanding distribution mismatch before intervention
  - Deciding whether IS correction is needed
  - A/B testing IS impact
- Controlled by `algorithm.rollout_is` flag (independent of weight
computation)

### 8. **Universal PPO Support**
- Integrated with **all PPO variants**: vanilla, GSPO, GPG, Clip-Cov,
KL-Cov, geo_mean
- Consistent interface across different policy loss functions
- Automatic weight application when enabled

---

## API and Configuration Changes

### Migration from Legacy TIS

#### ❌ **Before (REMOVED)**
```yaml
# Old TIS configuration - NO LONGER SUPPORTED
actor_rollout_ref:
  actor:
    tis_imp_ratio_cap: 2.0  # Removed from actor config
```

The legacy implementation:
- Only supported token-level truncation
- No metrics tracking
- Lacked numerical stability
- Limited configurability

#### ✅ **After (New Framework)**

Configuration moved to `algorithm` section for better organization:

```yaml
algorithm:
  # Main on/off switch: null = disabled, float = enabled
  rollout_is_threshold: 2.0

  # Control weight application (independent of metrics computation)
  rollout_is: true  # true = apply weights, false = metrics only

  # Optional: lower threshold (defaults to 1/upper if null)
  rollout_is_threshold_lower: null

  # Aggregation level: "token", "sequence", or "geometric"
  rollout_is_level: token

  # Bounding mode: "truncate" or "clip"
  rollout_is_mode: truncate

  # Veto threshold for catastrophic outliers (null = disabled)
  rollout_is_veto_threshold: 1e-4

# REQUIRED: Enable log probability calculation
actor_rollout_ref:
  rollout:
    calculate_log_probs: true
```

### Configuration Examples

**1. Token-level truncation (recommended starting point)**
```yaml
algorithm:
  rollout_is_threshold: 2.0
  rollout_is: true
  rollout_is_level: token
  rollout_is_mode: truncate
```

**2. Sequence-level clipping (more aggressive)**
```yaml
algorithm:
  rollout_is_threshold: 2.0
  rollout_is: true
  rollout_is_level: sequence
  rollout_is_mode: clip
```

**3. Metrics-only mode (monitoring without correction)**
```yaml
algorithm:
  rollout_is_threshold: 2.0
  rollout_is: false  # Compute metrics but don't apply weights
  rollout_is_level: token
  rollout_is_mode: truncate
```

**Example script:** `bash
examples/rollout_importance_sampling/run_with_rollout_is.sh`

---

## Code Changes Overview

### New Files (4 files, 1,442 lines)

1. **`verl/trainer/ppo/mismatch_helper.py`** (459 lines)
   - Core implementation of IS weight computation
   - Three aggregation levels: token, sequence, geometric
   - Two bounding modes: truncate, clip
   - Veto mechanism for outlier detection
   - Comprehensive metrics computation (IS + mismatch)
   - All computations in log-space for numerical stability
   - Memory-efficient design

2. **`docs/advance/rollout_is_migration.md`** (642 lines)
   - Comprehensive migration guide from legacy TIS
   - Detailed explanation of all configuration options
   - Recommended threshold ranges for each aggregation level
   - Troubleshooting guide and best practices
   - Metrics interpretation guide

3. **`examples/rollout_importance_sampling/README.md`** (242 lines)
   - Quick start guide with working examples
   - Configuration templates for common scenarios
   - Threshold tuning guidelines
   - Metrics monitoring instructions

4. **`examples/rollout_importance_sampling/run_with_rollout_is.sh`** (99
lines)
   - Complete working example script
   - Demonstrates token-level and sequence-level configurations
   - Ready to run with minimal modifications

### Modified Core Files (9 files)

1. **`verl/trainer/ppo/core_algos.py`** (~50 lines changed)
   - Removed legacy TIS logic (`tis_imp_ratio_cap`)
   - Added `rollout_is_weights` parameter to all policy loss functions
   - Unified IS weight application interface across all PPO variants:
     - `compute_policy_loss_vanilla`
     - `compute_policy_loss_gspo`
     - `compute_policy_loss_gpg`
     - `compute_policy_loss_clip_cov`
     - `compute_policy_loss_kl_cov`
     - `compute_policy_loss_geo_mean`
   - Special handling for `geo_mean` (sequence-level aggregation)

2. **`verl/trainer/ppo/ray_trainer.py`** (~52 lines added)
   - New method: `compute_rollout_importance_weights_and_add_to_batch()`
   - Centralized IS computation (once per batch, on driver)
- Conditional weight distribution to workers based on
`algorithm.rollout_is`
   - Metrics collection and aggregation
   - Integration with existing training loop

3. **`verl/trainer/config/algorithm.py`** (+18 lines)
   - Added 6 new Rollout IS parameters:
     - `rollout_is_threshold` (main on/off switch)
     - `rollout_is` (weight application control)
     - `rollout_is_threshold_lower`
     - `rollout_is_level`
     - `rollout_is_mode`
     - `rollout_is_veto_threshold`
   - Comprehensive docstrings explaining each parameter

4. **`verl/workers/config/actor.py`** (-1 line)
   - Removed deprecated `tis_imp_ratio_cap` parameter

5. **`verl/workers/actor/dp_actor.py`** (~26 lines changed)
   - Updated to use new `rollout_is_weights` parameter
   - Removed legacy TIS logic

6. **`verl/workers/actor/megatron_actor.py`** (~15 lines changed)
   - Updated to use new `rollout_is_weights` parameter
   - Removed legacy TIS logic

7. **Configuration Files** (4 files updated)
   - `verl/trainer/config/ppo_trainer.yaml`
   - `verl/trainer/config/ppo_megatron_trainer.yaml`
   - `verl/trainer/config/_generated_ppo_trainer.yaml`
   - `verl/trainer/config/_generated_ppo_megatron_trainer.yaml`
- Added default Rollout IS configuration section with explanatory
comments

### Testing (2 files, 530 lines)

1. **`tests/trainer/ppo/test_rollout_is.py`** (289 lines)
   - Unit tests for `mismatch_helper.py`
   - Coverage for all aggregation levels (token, sequence, geometric)
   - Coverage for all bounding modes (truncate, clip)
   - Veto mechanism tests
   - Edge case handling (zeros, extremes, empty sequences)
   - Numerical stability verification
   - Metrics correctness validation

2. **`tests/trainer/ppo/test_rollout_is_integration.py`** (241 lines)
   - Integration tests with PPO training loop
   - End-to-end workflow validation
   - Batch processing tests
   - Configuration validation
   - Metrics collection verification
   - Compatibility with distributed training

### Updated Recipes (2 files)

1. **`recipe/dapo/dapo_ray_trainer.py`** (+5 lines)
   - Updated imports to use new framework

2. **`recipe/dapo/run_dapo_qwen2.5_32b_tis.sh`** (~42 lines changed)
   - Migrated from legacy TIS to new Rollout IS configuration
   - Updated documentation and comments

### Documentation Updates (2 files)

1. **`docs/examples/config.rst`** (~22 lines changed)
   - Updated configuration examples
   - Added Rollout IS section

2. **`docs/index.rst`** (+1 line)
   - Added link to Rollout IS migration guide

---

## Implementation Highlights

### Centralized Architecture

The new design follows a clean separation of concerns:

```
ray_trainer.py (driver)
    └─> compute_rollout_importance_weights_and_add_to_batch()
         └─> mismatch_helper.compute_rollout_importance_weights()
              ├─> Computes IS weights (token/sequence/geometric)
              ├─> Applies bounding (truncate/clip)
              ├─> Veto mechanism for outliers
              ├─> Computes IS metrics
              └─> Computes mismatch metrics (KL, PPL)
    └─> Conditionally adds weights to batch (if rollout_is=True)
    └─> Distributes batch to workers

actor workers (dp_actor, megatron_actor)
    └─> Receive batch with rollout_is_weights (if enabled)
    └─> Pass weights to policy loss function

core_algos.py
    └─> All policy loss functions accept rollout_is_weights
    └─> Apply weights if provided: pg_losses *= rollout_is_weights
```

### Key Design Decisions

1. **Centralized Computation**: IS weights computed once on driver, not
per worker
   - Reduces redundant computation
   - Ensures consistency across workers
   - Simplifies debugging and metrics collection

2. **Configuration in Algorithm**: Moved from actor config to algorithm
config
- Better conceptual organization (algorithm-level concern, not
worker-level)
   - Easier to manage and validate
   - Consistent with other algorithm parameters

3. **Two-Level Control**:
   - `rollout_is_threshold`: Enables/disables entire system (null = off)
- `rollout_is`: Controls weight application (true = apply, false =
metrics only)
   - Allows flexible monitoring and gradual rollout

4. **Metrics Consolidation**: Mismatch metrics computed within IS weight
computation
   - Eliminates duplicate computation
   - Reduces memory overhead
   - Maintains metric accuracy

5. **Universal PPO Support**: Single interface for all PPO variants
   - Minimal code changes required
   - Consistent behavior across algorithms
   - Easy to add new variants

---

## Migration Guide

### For Users of Legacy TIS

**Step 1: Update your configuration file**

```yaml
# OLD (remove this)
actor_rollout_ref:
  actor:
    tis_imp_ratio_cap: 2.0

# NEW (add this)
algorithm:
  rollout_is_threshold: 2.0  # Use same value as old tis_imp_ratio_cap
  rollout_is: true
  rollout_is_level: token
  rollout_is_mode: truncate

# REQUIRED (add if not present)
actor_rollout_ref:
  rollout:
    calculate_log_probs: true
```

**Step 2: Monitor metrics**

The first time you run with the new configuration, check these metrics:
- `mismatch/rollout_is_eff_sample_size`: Should be > 80% of batch size
- `mismatch/rollout_is_veto_fraction`: Should be < 5%
- `mismatch/rollout_is_mean`: Should be close to 1.0

**Step 3: Tune if needed**

If effective sample size is too low:
- Increase `rollout_is_threshold`
- Try `rollout_is_mode: clip` with appropriate lower bound
- Consider `rollout_is_level: sequence` for more aggressive correction

For detailed guidance, see `docs/advance/rollout_is_migration.md`.

### For New Users

Start with recommended defaults:

```yaml
algorithm:
  rollout_is_threshold: 2.0
  rollout_is: true
  rollout_is_level: token
  rollout_is_mode: truncate

actor_rollout_ref:
  rollout:
    calculate_log_probs: true
```

Run the example script to see it in action:
```bash
bash examples/rollout_importance_sampling/run_with_rollout_is.sh
```

---

## Testing

### Unit Tests
- **289 lines** of comprehensive unit tests in `test_rollout_is.py`
- Covers all aggregation levels, bounding modes, and edge cases
- Validates numerical stability and correctness
- Fast execution (~1-2 seconds)

### Integration Tests
- **241 lines** of integration tests in `test_rollout_is_integration.py`
- End-to-end workflow with PPO training loop
- Distributed training compatibility
- Metrics collection validation
- Moderate execution time (~10-20 seconds)

### Running Tests
```bash
# Run all Rollout IS tests
pytest tests/trainer/ppo/test_rollout_is.py -v
pytest tests/trainer/ppo/test_rollout_is_integration.py -v

# Run specific test
pytest tests/trainer/ppo/test_rollout_is.py::test_token_level_truncate -v
```

---

## Metrics Reference

### Rollout IS Metrics (all prefixed with `mismatch/`)

| Metric | Description | Ideal Range |
|--------|-------------|-------------|
| `rollout_is_eff_sample_size` | Effective number of samples after IS |
> 80% of batch |
| `rollout_is_mean` | Mean IS weight | ~1.0 |
| `rollout_is_std` | Standard deviation of IS weights | Low variance |
| `rollout_is_p25` | 25th percentile | ~0.8-1.0 |
| `rollout_is_p50` | Median IS weight | ~1.0 |
| `rollout_is_p75` | 75th percentile | ~1.0-1.2 |
| `rollout_is_p95` | 95th percentile | < threshold |
| `rollout_is_p99` | 99th percentile | < threshold |
| `rollout_is_max` | Maximum weight | ≤ threshold |
| `rollout_is_min` | Minimum weight | ≥ lower threshold (clip mode) |
| `rollout_is_veto_fraction` | % sequences vetoed | < 5% |
| `rollout_is_catastrophic_token_fraction` | % catastrophic tokens | <
1% |
| `rollout_is_clipped_fraction` | % tokens clipped (clip mode) |
Variable |

### Mismatch Metrics (all prefixed with `mismatch/`)

| Metric | Description | What It Means |
|--------|-------------|---------------|
| `mismatch_kl` | Forward KL divergence | Distribution difference
(rollout vs training) |
| `mismatch_k3_kl` | K3 KL estimator | Stable KL estimate for small
divergences |
| `mismatch_training_ppl` | Training policy perplexity | Prediction
difficulty of training policy |
| `mismatch_rollout_ppl` | Rollout policy perplexity | Prediction
difficulty of rollout policy |
| `mismatch_ppl_ratio` | Ratio of training to rollout PPL | Relative
prediction difficulty |
| `mismatch_log_ppl_diff` | Log perplexity difference | Sequence-level
PPL mismatch |
| `mismatch_log_ppl_abs_diff` | Absolute log PPL difference | Magnitude
of mismatch |
| `mismatch_log_ppl_diff_max` | Max log PPL difference | Worst-case
mismatch |
| `mismatch_log_ppl_diff_min` | Min log PPL difference | Best-case
mismatch |
| `mismatch_training_log_ppl` | Log of training PPL | Log-scale training
perplexity |
| `mismatch_rollout_log_ppl` | Log of rollout PPL | Log-scale rollout
perplexity |

---

## Performance Impact

### Memory
- Minimal overhead: ~1-2% increase in peak memory usage
- Efficient log-space computation
- No large intermediate tensors

### Computation
- Negligible impact on training speed: < 1% overhead
- Centralized computation on driver (no per-worker redundancy)
- Optimized tensor operations

### Training Stability
- Significant improvement in stability when distribution mismatch exists
- Faster convergence in many scenarios
- Reduced risk of training collapse

---

## Breaking Changes

> [!IMPORTANT]
> This PR contains **BREAKING CHANGES** to the configuration API.

### Removed
- `actor_rollout_ref.actor.tis_imp_ratio_cap`: No longer supported

### Migration Required
All users of the legacy TIS implementation must update their
configuration files. See the migration guide above or
`docs/advance/rollout_is_migration.md` for detailed instructions.

### Backward Compatibility
- No backward compatibility with legacy TIS
- Configuration files with `tis_imp_ratio_cap` will raise validation
errors
- Affected recipes have been updated in this PR

---

## Pre-Submission Checklist

- [x] Search for similar PRs:
[https://github.com/volcengine/verl/pulls?q=is%3Apr+importance+sampling](https://github.com/volcengine/verl/pulls?q=is%3Apr+importance+sampling)
- [x] Format PR title as `[{modules}] {type}: {description}` (checked by
CI)
- **Suggested title:** `[BREAKING][rollout, trainer, algo] feat:
implement comprehensive Rollout Importance Sampling framework`
- [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)
- [x] Add/update
[documentation](https://github.com/volcengine/verl/tree/main/docs) (3
new docs, 2 updated)
- [x] Add unit and integration tests (530 lines of tests)
- [x] Once PR is ready for CI, send message in `ci-request` channel

---

## References

- **Blog post:** [When Speed Kills Stability: Demystifying RL Collapse
from the Inference-Training
Mismatch](https://yingru.notion.site/When-Speed-Kills-Stability-271211a558b7808d8b12d403fd15edda)
- **Migration guide:** `docs/advance/rollout_is_migration.md`
- **Examples:** `examples/rollout_importance_sampling/`
- **Tests:** `tests/trainer/ppo/test_rollout_is*.py`

---------

Co-authored-by: Yan Bai <bayan@nvidia.com>
 2025-10-13 17:05:29 +08:00
yangbaoxing
7f27789961 [fsdp,doc] refactor: rename warmup_style@FSDPOptimizerConfig -> lr_scheduler_type (#3739) 
### What does this PR do?

> Rename `warmup_style` in FSDPOptimizerConfig to `lr_scheduler_type` to
align with Hugging Face Trainer API。

The following pull request is for refactoring the optimizer, however,
the naming issue persists.
https://github.com/volcengine/verl/pull/3656 
### 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: ...
- [ ] 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: weiqi.li <weiqi.li@bytedance.com>
 2025-10-13 15:58:59 +08:00
lbk-sys
f50e5c2e8f [sglang] feat: add preparation for sglang+verl (#3506) 
### What does this PR do?
support npu for verl + sglang

```python
bash examples/grpo_trainer/run_qwen3_8b_grpo_sglang_1k_npu.sh
```


### Accuracy test
8b：
<img width="747" height="842" alt="8b"
src="https://github.com/user-attachments/assets/f36ef25a-b32f-4c76-97d0-2e5fe53ff183"
/>

30b：
<img width="759" height="850" alt="30b"
src="https://github.com/user-attachments/assets/97979002-7ebf-47fa-ae57-3e9b6637f12c"
/>

### Test


### 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).)

---------

Signed-off-by: lbk-sys <hello_lbk@163.com>
Co-authored-by: 1StepForever <wangww1Step@foxmail.com>
 2025-09-29 10:21:01 +08:00
CedricHuang
4ff3ce2fed [algo, perf] feat: Vectorize GRPO Advantage Estimator - 13～26x Speedup (#3635) 
### What does this PR do?  

Implements a vectorized GRPO advantage path for outcome-only RL in
core_algos.py, keeping the original implementation intact and
selectable. This yields large speedups at medium–large batch sizes by
replacing Python-side grouping loops with segment reductions and
one-pass gathers.


Results (CPU, Apple M-series example; float32):
```shell
[CPU] bs=  512 T= 512 G= 10 | orig=5.47ms vec=0.21ms speedup=26.16x
[CPU] bs= 1024 T=1024 G= 16 | orig=11.05ms vec=0.54ms speedup=20.60x
[CPU] bs= 2048 T=2048 G= 32 | orig=23.20ms vec=1.74ms speedup=13.32x
```

```shell
[GRPO] seed=0 groups=5 shape=torch.Size([64, 128]) mask_tokens=4147 adv_max_diff=2.384e-07 ret_max_diff=2.384e-07
[GRPO] seed=1 groups=8 shape=torch.Size([128, 256]) mask_tokens=16364 adv_max_diff=2.384e-07 ret_max_diff=2.384e-07
[GRPO] seed=2 groups=10 shape=torch.Size([512, 512]) mask_tokens=130968 adv_max_diff=4.768e-07 ret_max_diff=4.768e-07
```


### Checklist Before Starting

- [x] Search for similar PRs. Paste at least one query link here: #3634
- [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).)
 2025-09-27 17:21:08 +08:00
Chi Zhang
90154aeeb6 [doc] fix: fix doc (#3614) 
### What does this PR do?

- Fix url

### 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).)
 2025-09-25 16:11:43 +08:00
Chi Zhang
3dfa28ae32 [doc] feat: add model engine doc (#3611) 
### What does this PR do?

- Add model engine doc

### 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).)
 2025-09-25 14:25:44 +08:00
Shirley Wu
25d78fa913 [recipe] feat: CollabLLM integration for multiturn training (#3574) 
### What does this PR do?

This PR add [CollabLLM](https://aka.ms/CollabLLM) as a training recipe.
The added components include
- A customized `CollabLLMRewardManager` inheriting from
`AbstractRewardManager` to compute multiturn-aware rewards.
- A customized `CollabLLMAgentLoop` inheriting from `AgentLoop` to
sample future conversations with simulated users, which imports
`CollabLLMInteraction` from `verl/interactions/collabllm_interation.py`.

### 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.

The training rewards when running `train_rl_collabllm.sh` is increasing
in a relatively stable manner (on 8xH200):
<img width="964" height="480" alt="9baeb0700e3fa6a56596e14a54bc1049"
src="https://github.com/user-attachments/assets/53a810d8-1dd7-4145-bb28-4e475e9d7d9d"
/>

Validation reward:
<img width="974" height="538" alt="39364fd10523b0fde13d48645809f5e3"
src="https://github.com/user-attachments/assets/c34fe9e7-3d83-4132-8e1a-67e82c221d09"
/>

#### Samples of model generation
After training, when user asks generic questions with missing
information, the model learns to ask for clarification
<img width="1213" height="562" alt="c8e0ab31948a48ca396c7eccddd13673"
src="https://github.com/user-attachments/assets/ae41cd77-3c77-4402-b9d3-21993b046a18"
/>
and give suggestions:
<img width="1534" height="190" alt="7adb7d33eb9120d337c2a249c6a2dd22"
src="https://github.com/user-attachments/assets/84e1d8c1-f954-403f-b931-bce45cff1612"
/>

(In contrast, with the same prompt, **GPT-5** doesn't ask for any
clarification:)
<img width="1754" height="1126" alt="be8d8577584c0b2356cb352d6f294205"
src="https://github.com/user-attachments/assets/9b734848-9ed0-4496-af11-68bb8f8d8e08"
/>


### API and Usage Example

> Demonstrate how the API changes if any, and provide usage example(s)
if possible.

```python
# No change on the existing APIs
```

### Design & Code Changes

> Demonstrate the high-level design if this PR is complex, and list the
specific changes.

Changes:
- Main files under `recipe/collabllm`
- Registered `CollabLLMRewardManager` in
`workers/reward_manager/collabllm.py`
- Added `CollabLLMInteraction` in
`verl/interactions/collabllm_interation.py`

### 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). Added
to `verl/docs/algo/collabllm.md`.
- [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: The scripts
`train_rl_collabllm.sh` and `train_sft_collabllm.sh` are tested multiple
times.
- [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).)

---------

Co-authored-by: Chen Haiquan <chenhaiquan@bytedance.com>
 2025-09-25 09:53:39 +08:00
EduardDurech
26a734e740 [algo, perf] feat: Vectorize RLOO Advantage Estimator - 20x Speedup (#3555) 
Vectorize RLOO advantage estimator
130ms -> 6ms
Similar method can be done for other advantage estimators, I just don't
have time

Implements

$$r_i - \frac{\sum_{j\ne i} r_j}{G-1} = \frac{(G-1)r_i - \sum_{j\ne i}
r_j}{G-1} = \frac{G r_i - \sum_{j\in g} r_j}{G-1}$$

<img width="2199" height="628" alt="image"
src="https://github.com/user-attachments/assets/339e5bd2-6949-4460-a297-34268ffc1764"
/>
 2025-09-24 17:36:41 +08:00
OC
1b4af4440f [doc] fix: add faq doc to avoid vllm issue 22103 (#3595) 
### What does this PR do?

Provide a workaround for [vllm issue
22103](https://github.com/vllm-project/vllm/issues/22103), which may
result grad norm explosion.

You may hit this issue when match below conditions:
1. Using non-Hopper architecture GPUs, such as A100, L20, B200, etc.
2. Using vLLM as the inference engine.
3. The input and output texts are very long, for example, in multi-turn
scenarios using reasioning models like Qwen3 for RL training.

<img width="405" height="278" alt="截屏2025-09-24 下午1 37 50"
src="https://github.com/user-attachments/assets/47aec2e7-7c31-4cba-9f86-03af4f795457"
/>

The issue can be confirmed from comparing rollout_probs_diff_mean
metrics:

<img width="414" height="267" alt="截屏2025-09-24 下午1 39 24"
src="https://github.com/user-attachments/assets/f9cd484e-552a-49a4-b2b3-abb9e311c759"
/>

The workaround is:

`+actor_rollout_ref.rollout.engine_kwargs.vllm.disable_cascade_attn=True`

### 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

n/a

### API and Usage Example

n/a

### Design & Code Changes

n/a

### 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).)
 2025-09-24 13:47:36 +08:00
Yan Bai
bcd227598e [megatron] chore: add a docker image for with mcore0.15 and TE2.7 (#3540) 2025-09-22 10:59:33 +08:00
Zhen
c7922f0297 [doc] chore: Update ascend quick start document (#3527) 
### What does this PR do?

Remove reward maes, loss mae, total time ratio and throughput
information in ascend quick start document.

### 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

Not related.

### API and Usage Example

Not related.

### Design & Code Changes

Not related.

### 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).)
 2025-09-19 10:21:49 +08:00
X. HU
558d4dd581 [doc] fix: Update Qwen3-30B-A3B info in ascend_quick_start.rst (#3514) 
### What does this PR do?

1. Update the model name based on the training script to keep it
consistent with the Hugging Face official website.
https://github.com/volcengine/verl/pull/3189
2. Supplement the Qwen3-30B-A3B model info with actor.strategy as
megatron according to https://github.com/volcengine/verl/pull/3203


### 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).
- [ ] 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).)
 2025-09-19 09:57:42 +08:00
xvxuopop
f4e2047074 [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).)
 2025-09-17 18:48:24 +08:00
X. HU
2061894891 [model] feat: add qwen3 grpo 8b/32b script on ASCEND NPU (#3310) 
### What does this PR do?

add examples/grpo_trainer/run_qwen3_32b_npu.sh
<img width="1014" height="1111" alt="image"
src="https://github.com/user-attachments/assets/8cd59fc2-5f6a-419e-87ac-bf35a71856fb"
/>

add examples/grpo_trainer/run_qwen3_8b_npu.sh
<img width="844" height="930" alt="image"
src="https://github.com/user-attachments/assets/5c23c7a4-8729-4007-8828-027a8cda4779"
/>



> 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`, `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).
- [ ] 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: ...

> already support in https://github.com/volcengine/verl/pull/3300

- [ ] 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: ZLiao <a627465478@gmail.com>
Co-authored-by: ZLiao <a627465478@gmail.com>
 2025-09-15 10:13:01 +08:00
Huazhong
b86cd96eb7 [trainer, fsdp, megatron] feat: Support one step off async rl on Ascend NPU (#2924) 
### What does this PR do?

Since Ray's collective communication interface does not support the hccl
backend, we refer to the [example
code](https://docs.vllm.ai/en/latest/examples/offline_inference/rlhf.html)
of vLLM and complete the weight synchronization between actor and
rollout. This PR mainly introduces two changes:
1. Use `StatelessProcessGroup` and `PyNcclCommunicator` instead of ray's
`create_collective_group` to create weight synchronization communication
groups.
2. Use the `ray.get_runtime_context().get_accelerator_ids` API instead
of the environment variable `RAY_LOCAL_RANK` to set device in scenarios
where is_ray_noset_visible_devices is true, so as to fix the issue at
https://github.com/volcengine/verl/issues/2971.

### 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: ...
- [ ] 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).)
 2025-09-12 19:18:36 +08:00
OC
4c7812c40b [doc] fix: table column in document (#3430) 
### What does this PR do?

Added a missing column in Qwen3-30B-A3B MOE part.

### 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

n/a

### API and Usage Example

n/a

### Design & Code Changes

n/a
### 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).)
 2025-09-10 13:14:25 +08:00
X. HU
dfa3933ac4 [tool] feat: support local gsm8k dataset in example/data_preprocess (#3362) 2025-09-09 22:29:56 +08:00
Alex Kim
10054da277 [doc] fix: fix typo in skypilot_examples.rst (#3368) 2025-09-06 07:41:51 +08:00
Chi Zhang
90acc8abc1 [doc] fix: Update skypilot_examples.rst (#3344) 
### What does this PR do?

As title

### 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).)
 2025-09-04 19:01:59 +08:00
Alex Kim
f356fc1e56 [deployment, doc] feat: Add SkyPilot integration examples (#3333) 
### What does this PR do?

Adds SkyPilot integration examples for running verl training jobs
on Kubernetes/cloud platforms with GPUs. Includes configurations
for PPO, GRPO, and multi-turn tool usage training.

### Checklist Before Starting

- [x] Search for similar PRs. Paste at least one query link here:
https://github.com/volcengine/verl/pulls?q=is%3Apr+skypilot
- [x] Format the PR title as `[{modules}] {type}: {description}`

### Test

Validated SkyPilot YAML configurations for Ray cluster
initialization, dataset downloading, and distributed training setup with
H100 GPUs.

### API and Usage Example

```bash
# Launch PPO training on 2 nodes
sky launch -c verl-ppo examples/skypilot/verl-ppo.yaml --secret WANDB_API_KEY -y

# Launch GRPO training
sky launch -c verl-grpo examples/skypilot/verl-grpo.yaml --secret WANDB_API_KEY -y

# Launch multi-turn tool usage training
sky launch -c verl-multiturn examples/skypilot/verl-multiturn-tools.yaml --secret WANDB_API_KEY --secret HF_TOKEN -y
```

Design & Code Changes

- Added 3 SkyPilot YAML configurations for PPO, GRPO, and
multi-turn training
- Added `examples/skypilot/README.md` with setup guide
- Added `docs/examples/skypilot_examples.rst` documentation
- Updated `docs/index.rst` and `docs/start/multinode.rst` with
references

### 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).
- [ ] 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).)
 2025-09-04 16:56:00 +08:00
Chunyu
abe5e719ee [perf] feat: add npu silu &expand the scope of patch models (#3260) 
### What does this PR do?

- Add npu optimized silu.
- Patch silu and RMSNorm for more models.
- Refresh the performance of Qwen3-8B PEFT SFT.

### 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`
- [ ] 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).)
 2025-09-02 16:35:53 +08:00
Ethan (Yusheng) Su
fd1a121324 [hardware] fix: update source in dockerfile.rocm (#3284) 
### What does this PR do?

> Update the resource in `Dockerfile.rocm`

### 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

> docker build -f Dockerfile.rocm -t verl-rocm:local .
```
docker run --rm -it verl-rocm:local python -c "import torch; print('ok')"
```

### Design & Code Changes

> Update the resource in `Dockerfile.rocm`

### 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).)
 2025-09-01 11:32:44 +08:00
Feng Yao
b8dc5377c6 [BREAKING][vllm, fsdp] feat: add Rollout-Training Mismatch Fix -- Truncated importance sampling (#2953) 
### What does this PR do?

Support [vLLM-FSDP off-policy importance sampling
correction](https://fengyao.notion.site/off-policy-rl) using Truncated
Importance Sampling (TIS):

<img width="859" height="382" alt="TIS"
src="https://github.com/user-attachments/assets/adc8f797-aa14-4b29-b265-a682c281d08e"
/>




### 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
python3 -m verl.trainer.main_ppo \
    algorithm.adv_estimator=gae \
    data.train_files="$train_files" \
    data.val_files="$test_files" \
    data.train_batch_size=1024 \
    data.max_prompt_length=1024 \
    data.max_response_length=1024 \
    data.filter_overlong_prompts=True \
    data.truncation='error' \
    actor_rollout_ref.model.path=Qwen/Qwen2.5-32B-Instruct \
    actor_rollout_ref.model.enable_gradient_checkpointing=False \
    actor_rollout_ref.actor.optim.lr=1e-6 \
    actor_rollout_ref.model.use_remove_padding=True \
    actor_rollout_ref.actor.ppo_mini_batch_size=256 \
    actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 \
    actor_rollout_ref.model.enable_gradient_checkpointing=True \
    actor_rollout_ref.actor.fsdp_config.param_offload=False \
    actor_rollout_ref.actor.fsdp_config.optimizer_offload=False \
    actor_rollout_ref.actor.use_kl_loss=False \
    actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 \
    actor_rollout_ref.rollout.tensor_model_parallel_size=4 \
    actor_rollout_ref.rollout.name=vllm \
    actor_rollout_ref.rollout.gpu_memory_utilization=0.5 \
    critic.optim.lr=1e-5 \
    critic.model.use_remove_padding=True \
    critic.model.path=Qwen/Qwen2.5-32B-Instruct \
    critic.model.enable_gradient_checkpointing=False \
    critic.ppo_micro_batch_size_per_gpu=8 \
    critic.model.fsdp_config.param_offload=False \
    critic.model.fsdp_config.optimizer_offload=False \
    algorithm.use_kl_in_reward=False \
    trainer.critic_warmup=0 \
    trainer.logger='["console","wandb"]' \
    trainer.project_name='verl_example' \
    trainer.experiment_name='Qwen2.5-32B-Instruct_function_rm' \
    trainer.n_gpus_per_node=8 \
    trainer.nnodes=4 \
    trainer.save_freq=20 \
    trainer.test_freq=10 \
    trainer.total_epochs=15 \
    actor_rollout_ref.rollout.calculate_log_probs=True \   # add this config to return rollout prob
    +actor_rollout_ref.actor.behav_imp_weight_cap=10.0$@   # add this config to set up C value in TIS
```

### 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).)

---------

Co-authored-by: Narsil-Dinghuai Zhang 张鼎怀 <dinghuai233@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: LiyuanLucasLiu <llychinalz@gmail.com>
 2025-08-26 14:06:07 -07:00
Huapeng Zhou
27b63c724a [env, sglang] feat: Bump new sglang version to fix vlm OOM (#3216) 
### What does this PR do?
- Bump new version of sglang
- This version's sglang can fix vlm OOM issue, detail are in:
https://github.com/sgl-project/sglang/issues/9365

### Test

Using instruction following
https://github.com/zhaochenyang20/Awesome-ML-SYS-Tutorial/blob/main/rlhf/verl/multi-turn/release_log/latest_sglang.md

Now we have new version of sglang:
<img width="786" height="154" alt="image"
src="https://github.com/user-attachments/assets/bcec557e-196c-40c0-aa0f-c19d9f5c3e98"
/>

`gsm8k`:
using `verl/examples/sglang_multiturn/run_qwen2.5-3b_gsm8k_multiturn.sh`

[Wandb](https://wandb.ai/popsoda-university-of-washington/multi-turn-grpo-qwen2.5-3b-sglang/runs/dtcdin9b?nw=nwuserpopsoda)
<img width="532" height="329" alt="image"
src="https://github.com/user-attachments/assets/12f67d1a-a57e-497d-bfe5-6ff8c642e83f"
/>

It can work well.

### 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).)
 2025-08-26 13:29:36 +08:00
Liwei Ma
b4a410197c [doc] fix: fix a documentation typo for nsys (#3214) 
### What does this PR do?

[doc] fix: fix a documentation typo for nsys

### 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

### 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).)
 2025-08-26 10:11:15 +08:00
Blue Space
9b6a07fa77 [docker] feat: update to vllm 0.10.0, mcore 0.13, transformers 4.55.4 (#3192) 2025-08-26 05:17:57 +08:00
Shangwei-Li
2398d36be3 [recipe] feat: Add Qwen3 30B MoE NPU recipe (#3189) 
### What does this PR do?

> Update recipe/dapo/run_dapo_qwen3_30b_npu.sh.

### Checklist Before Starting

- [x] Search for similar PRs. Paste at least one query link here: 
https://github.com/volcengine/verl/pulls?q=fsdp+npu+30b+recipe
- [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.

Critic/rewards/mean Comparison Chart, where the orange line represents
ascend NPU, the pink line represents GPU.
<img width="3182" height="1272" alt="image"
src="https://github.com/user-attachments/assets/5c275127-6cb3-4bf9-ac89-0fa6abb668c0"
/>


### API and Usage Example

> Demonstrate how the API changes if any, and provide usage example(s)
if possible.

```shell
# Add code snippet or script demonstrating how to use this
cd /path/to/verl
bash recipe/dapo/run_dapo_qwen3_30b_base_npu_fsdp.sh
```

### 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).)

Co-authored-by: Shangwei-Li <lishangwei2@huawei.com>
 2025-08-25 19:38:23 +08:00
OC
f6f910069b [doc] fix: add qwen3moe-30b script and fix error in qwen3-235b (#3174) 
1. add qwen3moe-30b script for 1 to 4 H20 nodes with best performance
2. fix error in qwen3-235b: 
  - vllm enable_expert_parallel may result invalid output
  - megratron num_layers_in_last_pipeline_stage is a depreciate option
 
---------
Co-authored-by: Yan Bai <bayan@nvidia.com>
 2025-08-22 13:59:24 +08:00
ZLiao097
d2126e7afd [recipe] feat: support qwen2.5-32B DAPO training script on ASCEND NPU (#3146) 
### What does this PR do?
Provide an script for DAPO-training qwen2.5-32B on NPU, and update
experiment result.

### Checklist Before Starting

- [x] Search for similar PRs. Paste at least one query link here:
[[recipe] feat: support qwen3-8B/14B DAPO training on ASCEND
NPU](https://github.com/volcengine/verl/pull/2836)
- [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.

The following are some comparison charts of relevant data from script
testing, where red represents NPU and blue represents GPU.

Critic/rewards/mean Comparison Chart
<img width="1314" height="714" alt="image"
src="https://github.com/user-attachments/assets/3c303100-7106-491b-a6ea-e0bd1926076c"
/>

Response_length/mean Comparison Chart
<img width="1322" height="714" alt="image"
src="https://github.com/user-attachments/assets/9fa01f6f-2774-4b07-a38b-71cb6b5c8359"
/>

Val-core/math_dapo/acc/mean@32 Comparison Chart (Test by aime-2024)
<img width="1320" height="716" alt="image"
src="https://github.com/user-attachments/assets/b6912e3c-89c6-4999-90bb-fa961edc6e4a"
/>


### API and Usage Example

> Demonstrate how the API changes if any, and provide usage example(s)
if possible.

```bash
cd /path/to/verl
bash recipe/dapo/run_dapo_qwen2.5_32b_npu.sh
```

### 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).)

---------

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
 2025-08-20 20:13:57 +08:00
none0663
9d66534060 [doc] feat: documentation Update, Ray Job Management Commands (#3131) 
### What does this PR do?

Added two Ray CLI commands to the documentation for better job
monitoring:

1. **Job ID Retrieval Command**  
`ray job list | grep submission_id | grep JobStatus | grep RUNNING |
grep -oP 'raysubmit_[^'\''"]+' | head -n 1`
   This pipeline fetches the latest running job's submission ID by:
   - Filtering active jobs (`RUNNING` status)
   - Extracting `raysubmit_*` IDs
   - Returning the first match

2. **Continuous Log Streaming**  
   `ray job logs <Submission ID> --follow`  
Added the `--follow` parameter to enable real-time log streaming,
allowing users to:
   - Continuously monitor job output
   - Debug long-running processes interactively
   - Maintain persistent log connection until job completion

These additions enhance operational visibility for Ray job management
workflows.
 2025-08-20 11:03:48 +08:00
Zhe Zhang
dd13051602 Fix python version (#3103) 2025-08-18 20:46:09 -07:00
Blue Space
603c07d999 [doc, perf] feat: add profiling doc (#3113) 2025-08-19 09:06:33 +08:00
YumiMom
97b65c63c7 [perf] fix: fix npu profiler and add mstx UT (#3052) 
### What does this PR do?

- fix the parameter passing error for profile_level
- fix the error when creating npu profiler in discrete mode
- modify the execution script
- modify ascend profiling doc
- add the discrete parameter in tool_config
- add mstx_profile UT

### 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).)
 2025-08-18 15:18:08 +08:00
Chi Zhang
bc1b760fb1 [BREAKING] [rollout] feat: add a separate rollout worker (#3071) 
### What does this PR do?

- Introduce a separate rolloutworker that can be instantiated without
hybridengine
- Introduce a ModelConfig that wraps all model related config
- Remove hf_rollout (will replace with TP support in the future if
needed)
- Next PR: modify MegatronWorker to use separate rollout worker

### 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.

- [ ] 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).)

---------

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
 2025-08-18 10:57:33 +08:00
kang sheng
e31a883121 [rollout] fix: vllm sleep level=2 bug (#3082) 
### What does this PR do?

1. vllm sleep level=2 has bug and has been fixed:
https://github.com/vllm-project/vllm/pull/16889 and the bug fixed is
released in version 0.8.5:
https://github.com/vllm-project/vllm/releases/tag/v0.8.5
2. fix a typo in deepseek benchmark doc.
 2025-08-16 08:19:06 +08:00
Chunyu
28f6e4af7e [doc]fix: optimize ascend docs (#3063) 
### What does this PR do?

- 修复ascend_quick_start.rst中一些依赖软件的版本匹配错误。
- 支持现状表格中增加对actor.strategy和rollout.name的说明。
- 重命名ascend_profiling_en.rst和ascend_profiling_zh.rst，使文档标题看起来更美观些。 
<img width="402" height="103" alt="image"
src="https://github.com/user-attachments/assets/8f9ece22-315e-4f80-8157-04838f7467a3"
/>

### 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`
- [ ] 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: ...
- [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).)
 2025-08-15 13:24:21 +08:00
kang sheng
bd756c15c8 [BREAKING][rollout] feat: allow users pass all vllm/sglang engine args (#3037) 
This PR allows users to pass all vllm/sglang engine args and optimizes
qwen3 rollout speed through vllm Engine argument.

1. deprecate the default value of previous engine_kwargs
2. pass all the engine_kwargs to vllm/sglang engine
3. optimize Qwen3-235B rollout speed by setting TP=8 and enabling expert
parallel.

From top to bottom: tp=16 without EP, tp=8 without EP and tp=8 with EP.
<img width="1000" height="808" alt="image"
src="https://github.com/user-attachments/assets/6b096be4-3896-4e96-8916-d8d6e13a58cc"
/>

PS: The DeepSeek-V3's rollout slows down after enabling expert
parallelism.
 2025-08-14 19:12:26 +08:00
kang sheng
e2e4c35ecb [doc] feat: add benchmark for deepseek (#3046) 
Add a benchmark result for Deepseek. Other benchmark results are on the
way.
 2025-08-14 13:29:28 +08:00
kang sheng
65c59c719c [trainer,rollout,doc] feat: reduce minimum gpus to 96 for deepseek-v3 (#3019) 
### What does this PR do?
reduce minimum gpus to 96 for deepseek-v3 and 32 for Qwen3-235B

change details:
1. use cpu adam to save GPU memory
2. change vllm sleep level to 2 to save CPU memory
3. fix conflict between megatron HybridDeviceOptimizer and verl
load_megatron_optimizer.
4. provide new training scripts and document.

training logs:
DeepSeek-V3 with 12 Nodes:
<img width="3420" height="1308" alt="image"
src="https://github.com/user-attachments/assets/23bec729-bf39-41c8-a4c2-c51f389d052c"
/>


Qwen3-235B with 4 Nodes:
<img width="3426" height="1380" alt="image"
src="https://github.com/user-attachments/assets/4eeacab4-833f-4409-b294-10bd51d0fde9"
/>

sleep1 vs sleep2 speed on Qwen2.5-7B:
level1 mean:8.26s level2 mean: 8.33s
<img width="698" height="638" alt="image"
src="https://github.com/user-attachments/assets/e3dcbb4b-f841-4d7c-b60f-40da8ffe6c42"
/>
 2025-08-13 10:52:45 +08:00
Chunyu
3cc7695f4c [hardware, recipe] chore: support retool sft &update peft sft perf on npu (#3000) 
### What does this PR do?

- Add time statistics for all train steps.
- Support retool sft on ascend npu.
- Update peft sft performance.

### 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).)

---------

Co-authored-by: Tonyztj <1445297443@qq.com>
 2025-08-13 10:51:46 +08:00
黄石
92cbc2f417 [misc] feat: Support trackio (#3017) 
### What does this PR do?

Support Trackio, a lightweight experiment tracking library from Hugging
Face.

Features are listed in https://huggingface.co/blog/trackio

### 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.

- [ ] 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).)
 2025-08-12 13:02:20 +08:00
Blue Space
b79263ad60 [perf] refactor: part 2 - Profiler ci test and fixes (#3001) 
### What does this PR do?

[perf] refactor part 2: Profiler ci test and fixes

### 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).)
 2025-08-12 08:59:39 +08:00