---
language:
- en
- zh
license: apache-2.0
base_model:
- Qwen/Qwen3.5-9B
- Jackrong/Qwopus3.5-9B-v3.5
tags:
- unsloth
- qwen
- qwen3.5
- reasoning
- chain-of-thought
- Dense
- vLLM
- SGLang
pipeline_tag: image-text-to-text
datasets:
- nohurry/Opus-4.6-Reasoning-3000x-filtered
---
# 🌟Qwopus3.5-9B-v3.5-INT4-FOEM

<div align="left">
  <a href=https://ojs.aaai.org/index.php/AAAI/article/view/40123 target="_blank"><img src=https://img.shields.io/badge/Official%20Site-333399.svg?logo=homepage height=22px></a>
  <a href=https://huggingface.co/Xingyu-Zheng/Qwopus3.5-9B-v3.5-INT4-FOEM target="_blank"><img src=https://img.shields.io/badge/%F0%9F%A4%97%20Models-d96902.svg height=22px></a>
  <a href=https://github.com/Xingyu-Zheng/FOEM target="_blank"><img src= https://img.shields.io/badge/Page-bb8a2e.svg?logo=github height=22px></a>
  <a href=https://arxiv.org/abs/2507.11017 target="_blank"><img src=https://img.shields.io/badge/Report-b5212f.svg?logo=arxiv height=22px></a>
</div>

This is an unofficial quantized version of Qwopus3.5-9B-v3.5.

### 🧠 Quantization Framework

[GPTQModel](https://github.com/ModelCloud/GPTQModel)

## 🗺️ Quantization Method

[FOEM (AAAI 2026)](https://ojs.aaai.org/index.php/AAAI/article/view/40123)

FOEM is an improved quantization method over GPTQ. The resulting model preserves the same inference structure as GPTQ, ensuring compatibility with existing deployment pipelines while achieving better accuracy.

### 📚 Calibration Dataset

We randomly sampled 512 examples from [nohurry/Opus-4.6-Reasoning-3000x-filtered](https://huggingface.co/datasets/nohurry/Opus-4.6-Reasoning-3000x-filtered).

## 📋 Usage Example

This model can be deployed using standard frameworks such as **vLLM**, just like other GPTQModel-quantized models.

Example evaluation command:

```bash
lm-eval --model vllm --model_args pretrained=models/gptqmodel/Qwopus3.5-9B-v3.5-INT4-FOEM,tensor_parallel_size=1,gpu_memory_utilization=0.45 --tasks wikitext --batch_size 1
```

## ⚠️ Limitations & Intended Use

*(Adapted from the original repository of Jackrong/Qwopus3.5-9B-v3.5)*

- Possible overfitting if scaling exceeds optimal regime  
- Reasoning may still exhibit instability in edge cases  
- Tool-calling performance depends on environment integration  
- Not all capabilities are fully benchmarked yet  

## 🙏 Acknowledgements
Special thanks to [Jackrong](https://huggingface.co/Jackrong) for providing the original model: [Qwopus3.5-9B-v3.5](https://huggingface.co/Jackrong/Qwopus3.5-9B-v3.5).

## 📖 Citation

If you use this model in your research or projects, please cite:

```bibtex
@misc{jackrong_qwopus35_9b_v35,
  title        = {Qwopus3.5-9B-v3.5},
  author       = {Jackrong},
  year         = {2026},
  publisher    = {Hugging Face}
}
```

```bibtex
@misc{qubitium2024gptqmodel,
  author = {ModelCloud.ai and qubitium@modelcloud.ai},
  title = {GPT-QModel},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/modelcloud/gptqmodel}},
  note = {Contact: qubitium@modelcloud.ai},
  year = {2024},
}
```

```bibtex
@inproceedings{zheng2026first,
  title={First-order error matters: Accurate compensation for quantized large language models},
  author={Zheng, Xingyu and Qin, Haotong and Li, Yuye and Chu, Haoran and Wang, Jiakai and Guo, Jinyang and Magno, Michele and Liu, Xianglong},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={40},
  number={34},
  pages={28883--28891},
  year={2026}
}
```