KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#developer-tools #typescript #claude-code #codex #dsh #dsh-plugin #ai-agent #ai-agents #deepseek-harness #agent-skills #macos #cli

EchoInk

★ 78 Open GitHub ↗

EchoInk-R1: Exploring Audio-Visual Reasoning in Multimodal LLMs via Reinforcement Learning (🔥The Exploration of R1 for General Audio-Visual Reasoning with Qwen2.5-Omni)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# EchoInk-R1: Exploring Audio-Visual Reasoning in Multimodal LLMs via Reinforcement Learning

[📄 Technical Report (arXiv)](https://arxiv.org/pdf/2505.04623) • [🤗 Model (EchoInk-R1-7B)](https://huggingface.co/harryhsing/EchoInk-R1-7B) • [🤗 Dataset (AVQA-R1-6K)](https://huggingface.co/datasets/harryhsing/OmniInstruct_V1_AVQA_R1)

## Overview

**EchoInk-R1** is the **first general framework for unified audio-visual reasoning via reinforcement learning**, built upon [Qwen2.5-Omni-7B](https://github.com/QwenLM/Qwen2.5-Omni) and optimized using **Group Relative Policy Optimization (GRPO)**. It supports structured reasoning over synchronized audio-image inputs through multiple-choice question answering.

We introduce **AVQA-R1-6K**, a dataset derived from [OmniInstruct-v1](https://huggingface.co/datasets/m-a-p/OmniInstruct_v1), comprising:
- **4,490 training** samples  
- **1,911 validation** samples
- Each sample includes a synchronized audio-image pair with a multiple-choice question and four options.

Beyond our core study, **EchoInk-R1** provides an extensible **RL fine-tuning framework for Qwen2.5-Omni**, enabling easy adaptation to new multimodal reasoning tasks with minimal modifications.

## Performance

**EchoInk-R1-7B** achieves **85.77%** accuracy on the AVQA-R1-6K validation set, surpassing the base **Qwen2.5-Omni-7B** model (**80.53%**) using only **562 RL steps**.

All code, models, and data are released to support transparency and reproducibility.

## News
- **[2025/05/08]** Released the [AVQA-R1-6K dataset](https://huggingface.co/datasets/harryhsing/OmniInstruct_V1_AVQA_R1), [EchoInk-R1-7B model](https://huggingface.co/harryhsing/EchoInk-R1-7B), full training & evaluation pipeline, and [technical report](https://arxiv.org/abs/2505.04623).

## Highlights
- Built on **Qwen2.5-Omni-7B** with **GRPO-based RL**
- Supports **audio**, **image**, **video**, and **text** modalities
- Provides a complete pipeline: **dataset**, **training**, and **evaluation**

## Reflective Reasoning: Aha Moments

During training, **EchoInk-R1** exhibits **reflective reasoning behaviors**, where it revisits initial assumptions and refines its responses under ambiguous multimodal cues. These *“aha moments”* reveal its capacity for belief revision and deeper cross-modal understanding.

## Learning Dynamics

- **Accuracy reward** steadily improves throughout training, indicating that GRPO effectively guides the model toward more accurate and reasoned outputs.
- **Completion length** exhibits a two-phase trend: an initial increase as the model explores elaborated reasoning, followed by a gradual decline toward more concise and efficient answers.
- **Format reward** converges rapidly, showing that the model quickly internalizes the required response structure.

## Setup & Installation

### Environment Setup

```bash
git clone https://github.com/HarryHsing/EchoInk
cd EchoInk

conda create -n echoink-r1 python=3.11
conda activate echoink-r1
bash setup.sh
```
> 📝 **Gibberish / Repetitive Output ([Issue #13](https://github.com/HarryHsing/EchoInk/issues/13)):** If the model outputs gibberish or repetitive character loops during training/rollout (especially on `transformers>=4.57` or when using SDPA), install this specific `transformers` commit to fix it: `https://github.com/huggingface/transformers/tree/42ef218b58de79415ab45377a1e8de8dca3929f0`
> 
### Download Dataset

To download and extract the AVQA-R1-6K dataset:

```bash
git lfs install
git clone https://huggingface.co/datasets/harryhsing/AVQA-R1-6K
cd AVQA-R1-6K
tar -xzvf AVQA_R1.tar.gz

📁 Dataset Structure
AVQA_R1/
├── train/
│   ├── audios/
│   ├── images/
│   └── omni_rl_format_train.json
├── valid/
│   ├── audios/
│   ├── images/
│   └── omni_rl_format_valid.json
```

## Training

### Download Qwen2.5-Omni-7B Model

First, download the base model: [Qwen2.5-Omni-7B](https://huggingface.co/Qwen/Qwen2.5-Omni-7B)

Modify `config.json` of `Qwen2.5-Omni-7B` to include `"hidden_size": 3584` at the root level.

### Launch GRPO Training

```bash
bash ./src/scripts/run_grpo_image_audio_avqa.sh
```
> 📝 Set `per_device_train_batch_size=1` as in previous R1-V setups  
> 📝 To use custom data, follow the JSON format in `./src/make_omniInstruct_r1_dataset.py` for audio–image or audio–video tasks.  
> 📝 See [Qwen2.5-Omni issue #205](https://github.com/QwenLM/Qwen2.5-Omni/issues/205) if you run into a dtype mismatch error.  
> ⚙️ Trained on 8×A100 (80G) GPUs; also supported on 4×A100 (80G).

## Evaluation

Evaluate on the AVQA-R1-6K validation set:

```bash
python ./src/omniInstruct-v1_eval_valid.py # Run the model on the validation set
python ./src/omniInstruct-v1_cal_metrics_valid.py # Compute accuracy
```

## Acknowledgements

We thank the open-source community. This work builds on [Qwen2.5-Omni](https://github.com/QwenLM/Qwen2.5-Omni), [Video-R1](https://github.com/tulerfeng/Video-R1), [Open-R1-Video](https://github.com/Wang-Xiaodong1899/Open-R1-Video), [R1-V](https://github.com/Deep-Agent/R1-V), and [DeepSeek-R1](https://github.com/deepseek-ai/DeepSeek-R1).

## Citation

If you find EchoInk-R1 useful, please cite:

```bibtex
@article{xing2025echoink,
      title={{EchoInk-R1}: Exploring Audio-Visual Reasoning in Multimodal {LLMs} via Reinforcement Learning}, 
      author={Zhenghao Xing and Xiaowei Hu and Chi-Wing Fu and Wenhai Wang and Jifeng Dai and Pheng-Ann Heng},
      year={2025},
      journal={arXiv preprint arXiv:2505.04623}
}
```

EchoInk

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README