---
language:
- en
base_model: Qwen/Qwen3.5-9B
tags:
- chat
library_name: transformers
license: apache-2.0
---
# CodeV-SVA-VL: Automated Hardware Assertion Generation from Natural Language Specifications and Waveform Diagrams

<div align="center"> 
  <a href="https://huggingface.co/wyt2000/CodeV-SVA-VL-27B"><img src="https://img.shields.io/static/v1?label=Model&message=HuggingFace&color=yellow"></a> &ensp;
  <a href="https://github.com/wyt2000/CodeV-SVA"><img src="https://img.shields.io/static/v1?label=Code&message=Github&color=blue"></a> &ensp;
</div>

## Introduction

We introduce CodeV-SVA-VL, a family of vision language models designed to translate natural-language verification properties and waveform diagrams into SystemVerilog Assertions (SVAs).

Open-Source Plan:

- Benchmark ✓
- Model ✓
- Evaluation code ✓
- Paper
- Dataset ✓
- Data synthesis and training code

## Evaluation Results

| Model                  |                | Wave2SVA         |             |
| :--------------------  | :---------:    | :----------:     | :---------: |
|                        |   Func.@1      |   Func.@16       |  Func.@32   |
|                        |                |                  |             |
| Qwen3.6-Plus           |    53.8        |     **89.4**     |  **94.0**   |
| GPT-5.4                |    <u>53.9</u> |     77.4         |    84.0     |
| GLM-5V                 |    50.9        |     82.8         |    88.0     |
|                        |                |                  |             |
| Kimi-K2.7-Code-1T-A32B |    49.6        |     <u>86.4</u>  |    90.0     |
| GLM-4.6V-106B-A12B     |    26.1        |     66.3         |    76.0     |
|                        |                |                  |             |
| Qwen3.5-8B             |    16.5        |     65.2         |    70.0     |
| Qwen3.5-27B            |    25.4        |     68.3         |    74.0     |
|                        |                |                  |             |
| CodeV-SVA-VL-9B        |    51.9        |     79.4         |    82.0     |
| CodeV-SVA-VL-27B       |    **54.9**    |     85.5         |    <u>92.0</u>     |

## Models

| Model | Download |
| -------- | -------- |
|    CodeV-SVA-VL-9B     |   [🤗HuggingFace](https://huggingface.co/wyt2000/CodeV-SVA-VL-9B)    |
|    CodeV-SVA-VL-27B     |   [🤗HuggingFace](https://huggingface.co/wyt2000/CodeV-SVA-VL-27B)    |

## Citation

```latex
@misc{wu2026qimengcodevsvatrainingspecializedllms,
      title={QiMeng-CodeV-SVA: Training Specialized LLMs for Hardware Assertion Generation via RTL-Grounded Bidirectional Data Synthesis}, 
      author={Yutong Wu and Chenrui Cao and Pengwei Jin and Di Huang and Rui Zhang and Xishan Zhang and Zidong Du and Qi Guo and Xing Hu},
      year={2026},
      eprint={2603.14239},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2603.14239}, 
}
```