README.md

# DAPO Open-Source Implementation

> Open-Source Algorithm Implementation & Expriement Running: [Yuxuan Tong](https://tongyx361.github.io/), [Guangming Sheng](https://hk.linkedin.com/in/guangming-sheng-b50640211)

> [!IMPORTANT]
> **🔥 News!!!**
> - [2025/03] We published the training record of [an early version of DAPO (w/o Token-level PG Loss & Dynamic Sampling)](./run_dapo_early_qwen2.5_32b.sh), achieving 44% on AIME 2024, in [W&B](https://wandb.ai/verl-org/DAPO%20Reproduction%20on%20verl?nw=u7n2j5sht28).

🏠 [Homepage](https://dapo-sia.github.io/) | 📝 [Paper](https://dapo-sia.github.io/static/pdf/dapo_paper.pdf) | 🤗 [Datasets&Models@HF](https://huggingface.co/collections/BytedTsinghua-SIA/dapo-67d7f1517ee33c8aed059da0) | 🐱 [Code@GitHub](https://github.com/volcengine/verl/tree/gm-tyx/puffin/main/recipe/dapo) | 🐱 [Repo@GitHub](https://github.com/BytedTsinghua-SIA/DAPO)

> We propose the **D**ecoupled Clip and Dynamic s**A**mpling **P**olicy **O**ptimization (DAPO) algorithm. By making our work publicly available, we provide the broader research community and society with practical access to scalable reinforcement learning, enabling all to benefit from these advancements. Our system is based on the awesome [verl](https://github.com/volcengine/verl) framework. Thanks for their great work! Applying DAPO training to Qwen2.5-32B base model proves to outperform the previous state-of-the-art DeepSeek-R1-Zero-Qwen-32B on AIME 2024, achieving **50%** accuracy with **50%** less training steps.
>
> ![dapo-main-result](https://dapo-sia.github.io/static/images/score.png)

## Quickstart

1. Prepare the datasets **on the Ray cluster**:

```bash
bash prepare_dapo_data.sh # This downloads the datasets to ${HOME}/verl/data by default
```

2. Submit the job to the Ray cluster **from any machine**:

```bash
cd verl # Repo root
export RAY_ADDRESS="http://${RAY_IP:-localhost}:8265" # The Ray cluster address to connect to
export WORKING_DIR="${PWD}" # The local directory to package to the Ray cluster
# Set the runtime environment like env vars and pip packages for the Ray cluster in yaml
export RUNTIME_ENV="./verl/trainer/runtime_env.yaml"
bash recipe/dapo/run_dapo_qwen2.5_32b.sh
```

## Reproduction Runs

| Setup | AIME 2024 Acc. | Training Script | Training Record |
|-------|----------------------|-----------------|-----------------|
| DAPO w/o Token-level PG Loss & Dynamic Sampling | 44% | [run_dapo_early_qwen2.5_32b.sh](./run_dapo_early_qwen2.5_32b.sh) | [W&B](https://wandb.ai/verl-org/DAPO%20Reproduction%20on%20verl?nw=u7n2j5sht28) |
| DAPO | 50% | [run_dapo_qwen2.5_32b.sh](./run_dapo_qwen2.5_32b.sh) | W&B (Coming soon) |

## Configuration

> [!NOTE]
> Most experiments in the paper, including the best-performant one, are run without Overlong Filtering because it's somehow overlapping with Overlong Reward Shaping in terms of properly learning from the longest outputs. So we don't implement it here.

### Separated Clip Epsilons (-> Clip-Higher)

An example configuration:

```yaml
actor_rollout_ref:
  actor:
    clip_ratio_low: 0.2
    clip_ratio_high: 0.28
```

`clip_ratio_low` and `clip_ratio_high` specify the $\varepsilon_{\text {low }}$ and $\varepsilon_{\text {high }}$ in the DAPO objective.

Core relevant code:

```python
pg_losses1 = -advantages * ratio
pg_losses2 = -advantages * torch.clamp(ratio, 1 - cliprange_low, 1 + cliprange_high)
pg_losses = torch.maximum(pg_losses1, pg_losses2)
```

### Dynamic Sampling (with Group Filtering)

An example configuration:

```yaml
data:
  gen_batch_size: 1536
  train_batch_size: 512
algorithm:
  filter_groups:
    enable: True
    metric: acc # score / seq_reward / seq_final_reward / ...
    max_num_gen_batches: 10 # Non-positive values mean no upper limit
```

Setting `filter_groups.enable` to `True` will filter out groups whose outputs' `metric` are all the same, e.g., for `acc`, groups whose outputs' accuracies are all 1 or 0.

The trainer will repeat sampling with `gen_batch_size` until there are enough qualified groups for `train_batch_size` or reaching the upper limit specified by `max_num_gen_batches`.

Core relevant code:

```python
prompt_bsz = self.config.data.train_batch_size
if num_prompt_in_batch < prompt_bsz:
    print(f'{num_prompt_in_batch=} < {prompt_bsz=}')
    num_gen_batches += 1
    max_num_gen_batches = self.config.algorithm.filter_groups.max_num_gen_batches
    if max_num_gen_batches <= 0 or num_gen_batches < max_num_gen_batches:
        print(f'{num_gen_batches=} < {max_num_gen_batches=}. Keep generating...')
        continue
    else:
        raise ValueError(
            f'{num_gen_batches=} >= {max_num_gen_batches=}. Generated too many. Please check your data.'
        )
else:
    # Align the batch
    traj_bsz = self.config.data.train_batch_size * self.config.actor_rollout_ref.rollout.n
    batch = batch[:traj_bsz]
```

### Flexible Loss Aggregation Mode (-> Token-level Policy Gradient Loss)

An example configuration:

```yaml
actor_rollout_ref:
  actor:
    loss_agg_mode: "token-mean" # / "seq-mean-token-sum" / "seq-mean-token-mean"
    # NOTE: "token-mean" is the default behavior
```

Setting `loss_agg_mode` to `token-mean` will mean the (policy gradient) loss across all the tokens in all the sequences in a mini-batch.

Core relevant code:

```python
if loss_agg_mode == "token-mean":
    pg_loss = verl_F.masked_mean(pg_losses, eos_mask)
elif loss_agg_mode == "seq-mean-token-sum":
    pg_loss = torch.sum(pg_losses * eos_mask, dim=-1) / torch.sum(eos_mask, dim=-1)
    pg_loss = torch.mean(pg_loss)
elif loss_agg_mode == "seq-mean-token-mean":
    pg_loss = torch.sum(pg_losses * eos_mask, dim=-1) / torch.sum(eos_mask, dim=-1)
    pg_loss = torch.mean(pg_loss)
else:
    raise ValueError(f"Invalid loss_agg_mode: {loss_agg_mode}")
```

### Overlong Reward Shaping

An example configuration:

```yaml
data:
  max_response_length: 20480 # 16384 + 4096
reward_model:
  overlong_buffer: 
    enable: True
    len: 4096
    penalty_factor: 1.0
```

Setting `overlong_buffer.enable` to `True` will penalize the outputs whose lengths are overlong but still within the hard context limit.

Specifically, the penalty increases linearly from `0` to `overlong_buffer.penalty_factor` when the length of the output exceeds the `max_response_length` by `0` to `overlong_buffer.len` tokens.

Core relevant code:

```python
if self.overlong_buffer_cfg.enable:
    overlong_buffer_len = self.overlong_buffer_cfg.len
    expected_len = self.max_resp_len - overlong_buffer_len
    exceed_len = valid_response_length - expected_len
    overlong_penalty_factor = self.overlong_buffer_cfg.penalty_factor
    overlong_reward = min(-exceed_len / overlong_buffer_len * overlong_penalty_factor, 0)
    reward += overlong_reward
```