|
|
--- |
|
|
base_model: Qwen/Qwen2.5-32B |
|
|
language: |
|
|
- en |
|
|
library_name: transformers |
|
|
license: apache-2.0 |
|
|
pipeline_tag: text-generation |
|
|
--- |
|
|
|
|
|
# K2-Think: A Parameter-Efficient Reasoning System |
|
|
|
|
|
📚 [Paper](https://huggingface.co/papers/2509.07604) - 📝 [Code](https://github.com/MBZUAI-IFM/K2-Think-SFT) - 🏢 [Project Page](https://k2think.ai) |
|
|
|
|
|
<center><img src="banner.png" alt="k2-think-banner"/></center> |
|
|
|
|
|
<br> |
|
|
|
|
|
K2-Think is a 32 billion parameter open-weights general reasoning model with strong performance in competitive mathematical problem solving. |
|
|
|
|
|
# Quickstart |
|
|
|
|
|
### Transformers |
|
|
You can use `K2-Think` with Transformers. If you use `transformers.pipeline`, it will apply the chat template automatically. If you use `model.generate` directly, you need to apply the chat template mannually. |
|
|
|
|
|
```python |
|
|
from transformers import pipeline |
|
|
import torch |
|
|
|
|
|
model_id = "LLM360/K2-Think" |
|
|
|
|
|
pipe = pipeline( |
|
|
"text-generation", |
|
|
model=model_id, |
|
|
torch_dtype="auto", |
|
|
device_map="auto", |
|
|
) |
|
|
|
|
|
messages = [ |
|
|
{"role": "user", "content": "what is the next prime number after 2600?"}, |
|
|
] |
|
|
|
|
|
outputs = pipe( |
|
|
messages, |
|
|
max_new_tokens=32768, |
|
|
) |
|
|
print(outputs[0]["generated_text"][-1]) |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
# Evaluation & Performance |
|
|
Detailed evaluation results are reported in out [Tech Report](https://k2think-about.pages.dev/assets/tech-report/K2-Think_Tech-Report.pdf) |
|
|
|
|
|
## Benchmarks (pass\@1, average over 16 runs) |
|
|
|
|
|
| Domain | Benchmark | K2-Think | |
|
|
| ------- | ---------------- | -----------: | |
|
|
| Math | AIME 2024 | 90.83 | |
|
|
| Math | AIME 2025 | 81.24 | |
|
|
| Math | HMMT 2025 | 73.75 | |
|
|
| Math | OMNI-Math-HARD | 60.73 | |
|
|
| Code | LiveCodeBench v5 | 63.97 | |
|
|
| Science | GPQA-Diamond | 71.08 | |
|
|
|
|
|
--- |
|
|
|
|
|
## Inference Speed |
|
|
|
|
|
We deploy K2-THINK on Cerebras Wafer-Scale Engine (WSE) systems, leveraging the world’s largest processor and speculative decoding to achieve unprecedented inference speeds for our 32B reasoning system. |
|
|
|
|
|
| Platform | Throughput (tokens/sec) | Example: 32k-token response (time) | |
|
|
| --------------------------------- | ----------------------: | ---------------------------------: | |
|
|
| **Cerebras WSE (our deployment)** | **\~2,000** | **\~16 s** | |
|
|
| Typical **H100/H200** GPU setup | \~200 | \~160 s | |
|
|
|
|
|
--- |
|
|
|
|
|
## Safety Evaluation |
|
|
|
|
|
Aggregated across four safety dimensions (**Safety-4**): |
|
|
|
|
|
| Aspect | Macro-Avg | |
|
|
| ------------------------------- | --------: | |
|
|
| High-Risk Content Refusal | 0.83 | |
|
|
| Conversational Robustness | 0.89 | |
|
|
| Cybersecurity & Data Protection | 0.56 | |
|
|
| Jailbreak Resistance | 0.72 | |
|
|
| **Safety-4 Macro (avg)** | **0.75** | |
|
|
|
|
|
--- |
|
|
# Terms of Use |
|
|
|
|
|
This model is released strictly for **research and educational purposes**. |
|
|
By downloading, using, or interacting with this model, you agree to the following conditions: |
|
|
|
|
|
1. **Research Only** |
|
|
The model is provided as part of an academic and research project. |
|
|
It is **not intended for commercial deployment** or production use without explicit permission. |
|
|
|
|
|
2. **Prohibited Uses** |
|
|
You may **not** use this model: |
|
|
- For any **illegal, unlawful, or harmful activities**, including but not limited to generating or disseminating malicious content, engaging in fraud, violating privacy, or spreading misinformation. |
|
|
- In applications that could directly cause **harm, injury, or safety risks** to individuals or society. |
|
|
|
|
|
3. **No Warranty** |
|
|
The model is provided **“as is” without warranties** of any kind. |
|
|
The authors and institutions involved bear no responsibility for consequences arising from its use. |
|
|
|
|
|
4. **Attribution** |
|
|
When using or referencing the model in research, publications, or derivative works, proper **citation and attribution** to the authors and project must be given. |
|
|
|
|
|
5. **Compliance** |
|
|
You are responsible for ensuring that your use of the model complies with all **applicable laws, regulations, and ethical guidelines** in your jurisdiction. |
|
|
|
|
|
--- |
|
|
|
|
|
# Citation |
|
|
|
|
|
```bibtex |
|
|
@misc{cheng2025k2thinkparameterefficientreasoning, |
|
|
title={K2-Think: A Parameter-Efficient Reasoning System}, |
|
|
author={Zhoujun Cheng and Richard Fan and Shibo Hao and Taylor W. Killian and Haonan Li and Suqi Sun and Hector Ren and Alexander Moreno and Daqian Zhang and Tianjun Zhong and Yuxin Xiong and Yuanzhe Hu and Yutao Xie and Xudong Han and Yuqi Wang and Varad Pimpalkhute and Yonghao Zhuang and Aaryamonvikram Singh and Xuezhi Liang and Anze Xie and Jianshu She and Desai Fan and Chengqian Gao and Liqun Ma and Mikhail Yurochkin and John Maggs and Xuezhe Ma and Guowei He and Zhiting Hu and Zhengzhong Liu and Eric P. Xing}, |
|
|
year={2025}, |
|
|
eprint={2509.07604}, |
|
|
archivePrefix={arXiv}, |
|
|
primaryClass={cs.LG}, |
|
|
url={https://arxiv.org/abs/2509.07604}, |
|
|
} |
|
|
``` |
