Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
The first Large Audio Language Model that enables native in-depth thinking, which is trained on large-scale audio Chain-of-Thought data.
| Date | Stars |
|---|---|
| 2026-07-31 | 297 |
| 2026-08-02 | 297 |
| 2026-08-06 | 297 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Audio-Reasoner
<p align="center">
<img src="assets\title.png" width="90%"/>
</p>
## Abstract
We implemented inference scaling on **Audio-Reasoner**, a large audio language model, enabling **deepthink** and **structured chain-of-thought (COT) reasoning** for multimodal understanding and reasoning. To achieve this, we constructed CoTA, a high-quality dataset with **1.2M reasoning-rich samples** using structured COT techniques. Audio-Reasoner achieves state-of-the-art results on **MMAU-mini(+25.42%)** and **AIR-Bench-Chat(+14.57%)** benchmarks.
<p align="center">
Audio-Reasoner-7B <a href="https://huggingface.co/zhifeixie/Audio-Reasoner/tree/main">🤗</a> | CoTA Dataset <a href="https://huggingface.co/datasets/zhifeixie/Audio-Reasoner-CoTA">🤗</a> <br>
Paper <a href="https://arxiv.org/abs/2503.02318"> 📑</a> ｜ Wechat <a href="https://github.com/xzf-thu/Audio-Reasoner/blob/main/assets/wechat.jpg">💭</a> ｜ Code <a href="https://github.com/xzf-thu/Audio-Reasoner"> ⚙️</a>
<br>
<a href="#demo"> Demo</a> • <a href="#install">Install</a> • <a href="#quick-start">Quick Start</a> • <a href="#faq">FAQ</a> • <a href="#contact">Contact us</a><br>
<br>
If you like us, pls give us a star⭐ !
</p>
## Main Results
<p align="center">
<img src="assets\main_result.png" width="100%"/>
</p>
## News and Updates
- **2025.03.05:** ✅**Audio-Reasoner-7B checkpoint is released on HuggingFace<a href="https://huggingface.co/zhifeixie/Audio-Reasoner/tree/main">🤗</a> !**
- **2025.03.05:** ✅**Audio-Reasoner Paper is uploaded to arXiv<a href="https://arxiv.org/abs/2503.02318"> 📑</a>.**
- **2025.03.04:** ✅**Demos, inference code and evaluation results have been released.**
- **2025.03.04:** ✅**Create this repo.**
## Roadmap
- **2025.03:** **🔜Upload CoTA dataset to HuggingFace🤗.**
- **2025.04:** **🔜Open-source data systhesis pipeline and training code**.
## Demo
<p align="center" width="80%">
<video controls src="https://github.com/user-attachments/assets/d50f75e7-288b-454b-92a3-c6f058be231b" title="v" width="100%"></video>
</p>
## Features
✅ Audio-Reasoner enables **deep reasoning and inference scaling** in audio-based tasks, built on Qwen2-Audio-Instruct with structured CoT training.
✅ CoTA offers **1.2M** high-quality captions and QA pairs across domains for structured reasoning and enhanced pretraining.
✅ Pretrained model and dataset encompassing various types of audio including sound, music, and speech, has achieved state-of-the-art results across multiple benchmarks. Refer to our <a href="https://arxiv.org/abs/2503.02318">paper</a> for details.
## Install
**Clone and install**
- Clone the repo
``` sh
git clone https://github.com/xzf-thu/Audio-Reasoner.git
cd Audio-Reasoner
```
- Install the required packages
```sh
conda create -n Audio-Reasoner python=3.10
conda activate Audio-Reasoner
pip install -r requirements.txt
pip install transformers==4.48.0
```
## Quick Start
**Chat using ms-swift**
```sh
import os
import re
from typing import List, Literal
from swift.llm import InferEngine, InferRequest, PtEngine, RequestConfig, load_dataset, get_template
from swift.plugin import InferStats
def infer_stream(engine: 'InferEngine', infer_request: 'InferRequest'):
request_config = RequestConfig(max_tokens=2048, temperature=0, stream=True)
metric = InferStats()
gen = engine.infer([infer_request], request_config, metrics=[metric])
query = infer_request.messages[0]['content']
output = ""
print(f'query: {query}\nresponse: ', end='')
for resp_list in gen:
if resp_list[0] is None:
continue
print(resp_list[0].choices[0].delta.content, end='', flush=True)
output += resp_list[0].choices[0].delta.content
print()
print(f'metric: {metric.compute()}')
return output
def get_message(audiopath, prompt):
messages = [
{"role": "system", "content": system},Excerpt of 7,005 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:34b2d832b13bd783, llm:description: 'The first Large Audio Language Model that enables native in-depth thinking, trained on large-scale audio Chain-of-Thought data.'
matched fp:34b2d832b13bd783, llm:description: 'The first Large Audio Language Model that enables native in-depth thinking, trained on large-scale audio Chain-of-Thought data.'
matched fp:34b2d832b13bd783, llm:description: 'The first Large Audio Language Model that enables native in-depth thinking, trained on large-scale audio Chain-of-Thought data.'
matched fp:34b2d832b13bd783, llm:description: 'The first Large Audio Language Model that enables native in-depth thinking, trained on large-scale audio Chain-of-Thought data.'