SAGEHLS

KO | EN

gitlite — search

#python #typescript #llm #android #machine-learning #deep-learning #linux #java #awesome #vue #pytorch #security #ios #open-source

(★ 9)

SAGE-HLS: Syntax-Aware AST-Guided LLM for High-Level Synthesis Code Generation

SAGEHLS 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# SAGE-HLS
SAGE-HLS: Syntax-Aware AST-Guided LLM for High-Level Synthesis Code Generation

This repository contains the resources, code, and documentation supporting our work on training a model for HLS-C++ code generation. Our work demonstrates how to leverage transformer-based architectures for understanding and generating HLS code, with a focus on applications using Xilinx Vitis HLS tool.

## Model Details

- **Model Name:** [mashnoor/hls-vitis-ast-qwen-coder-7B](https://huggingface.co/mashnoor/hls-vitis-ast-qwen-coder-7B)
- **Base Architecture:** Qwen2.5-Coder-7B
- **Training Framework:** [Unsloth](https://github.com/unslothai/unsloth)

## Dataset

- **Dataset:** [mashnoor/hls-ast-sagehls](https://huggingface.co/datasets/mashnoor/hls-ast-sagehls)
- The dataset consists of HLS code samples, annotated with AST information, curated to support training and evaluation of code language models in the hardware design domain.

## Quick Start

1. **Open and run the notebook**
   - Open `run_model.ipynb` in [Google Colab](https://colab.research.google.com/), JupyterLab, or any compatible environment with GPU support.
   - The notebook will:
     - Install Unsloth.
     - Load the fine-tuned Qwen Coder 7B model from Hugging Face.
     - Provide a prompt template for HLS C++ code generation.
     - Run inference to generate HLS code based on Verilog module descriptions.

2. **Example Usage**

The notebook includes a prompt template for translating Verilog module specifications into synthesizable HLS-C++ code. You can modify the prompt or input your own Verilog modules to generate corresponding HLS implementations.

```python
   # Example: Loading and running the model (see notebook for full details)
   from unsloth import FastLanguageModel
   model, tokenizer = FastLanguageModel.from_pretrained(
       model_name = "mashnoor/hls-vitis-ast-unverified-qwen-coder-7B",
       max_seq_length = 2048,
       load_in_4bit = True,
   )
   FastLanguageModel.for_inference(model)
   # ...see the notebook for prompt formatting and generation code
   ```

## Citation

If you use our model, dataset, or code, please cite our paper:

```bibtex
@misc{khan2025sagehlssyntaxawareastguidedllm,
      title={SAGE-HLS: Syntax-Aware AST-Guided LLM for High-Level Synthesis Code Generation}, 
      author={M Zafir Sadik Khan and Nowfel Mashnoor and Mohammad Akyash and Kimia Azar and Hadi Kamali},
      year={2025},
      eprint={2508.03558},
      archivePrefix={arXiv},
      primaryClass={cs.PL},
      url={https://arxiv.org/abs/2508.03558}, 
}
```

## License

This repository is released under the Apache License 2.0. See [LICENSE](LICENSE) for details.

## Acknowledgements

- [Unsloth](https://github.com/unslothai/unsloth) for efficient model fine-tuning.
- [Hugging Face](https://huggingface.co/) for model and dataset hosting.

SAGEHLS

SAGEHLS 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

SAGEHLS

SAGEHLS 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README