---
license: apache-2.0
language:
  - en
library_name: transformers
pipeline_tag: image-text-to-text
base_model: Feng613/SleepVLM-3B
base_model_relation: quantized
tags:
  - sleep-staging
  - polysomnography
  - PSG
  - explainable-AI
  - AASM
  - vision-language-model
  - medical
  - EEG
  - EOG
  - EMG
  - quantized
  - 4-bit
  - auto-round
  - rule-grounded
  - multimodal
datasets:
  - Feng613/MASS-EX
model-index:
  - name: SleepVLM-3B-W4A16
    results:
      - task:
          type: image-text-to-text
          name: Sleep Stage Classification
        dataset:
          name: MASS-SS1
          type: mass-ss1
        metrics:
          - name: Accuracy
            type: accuracy
            value: 0.827
          - name: Macro-F1
            type: macro-f1
            value: 0.788
          - name: Cohen's Kappa
            type: cohens-kappa
            value: 0.758
      - task:
          type: image-text-to-text
          name: Sleep Stage Classification
        dataset:
          name: ZUMS
          type: zums
        metrics:
          - name: Accuracy
            type: accuracy
            value: 0.798
          - name: Macro-F1
            type: macro-f1
            value: 0.751
          - name: Cohen's Kappa
            type: cohens-kappa
            value: 0.727
---

<div align="center">
<img src="SleepVLM_logo.png" alt="SleepVLM Logo" width="360">

# SleepVLM-3B-W4A16

### Quantized Version — Explainable and Rule-Grounded Sleep Staging via a Vision-Language Model

[Paper (coming soon)]() |
[GitHub](https://github.com/Deng-GuiFeng/SleepVLM) |
[Full-Precision Version](https://huggingface.co/Feng613/SleepVLM-3B) |
[MASS-EX Dataset](https://huggingface.co/datasets/Feng613/MASS-EX) |
[Collection](https://huggingface.co/collections/Feng613/sleepvlm)

</div>

---

> **Associated Paper:**
> Guifeng Deng, Pan Wang, Jiquan Wang, Tao Li, Haiteng Jiang. "SleepVLM: Explainable and Rule-Grounded Sleep Staging via a Vision-Language Model." *In preparation.*
> This repository will be made public upon release of the preprint.

## Overview

**SleepVLM-3B-W4A16** is the 4-bit weight-quantized version of [SleepVLM-3B](https://huggingface.co/Feng613/SleepVLM-3B), a rule-grounded vision-language model for explainable automated sleep staging from polysomnography (PSG) recordings. This quantized variant achieves 2.2x faster inference and 55% model size reduction with minimal performance degradation (kappa drop ≤1.6 pp), enabling deployment on a single consumer-grade GPU (e.g., NVIDIA RTX 4090, 24 GB).

The quantization was performed using [Intel AutoRound](https://github.com/intel/auto-round) (W4A16: 4-bit weights, 16-bit activations) on the language model layers only. The vision encoder and lm_head are retained in float16 precision.

For full details about the SleepVLM framework and training pipeline, see the [full-precision model card](https://huggingface.co/Feng613/SleepVLM-3B).

## Model Details

| Property | Value |
|----------|-------|
| Base model | [SleepVLM-3B](https://huggingface.co/Feng613/SleepVLM-3B) (fine-tuned from [Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct)) |
| Model size | 3.2 GB (vs 7.1 GB full-precision, **-54.9%**) |
| Inference speed | 4.15 epoch/s (vs 1.89 epoch/s, **+2.20x**) |
| Precision | W4A16 (4-bit weights, 16-bit activations) |
| Quantization method | [Intel AutoRound](https://github.com/intel/auto-round) v0.9.2 |
| Quantized layers | `model.language_model.layers` (36 transformer blocks) |
| Non-quantized layers | Vision encoder + lm_head (float16) |
| Group size | 128 |
| Calibration samples | 5,000 (stratified by sleep stage) |
| Input | Three consecutive 30-s PSG epoch images (448 x 224 px) |
| PSG channels | F4-M1, C4-M1, O2-M1, LOC, ROC, Chin EMG |

## Intended Use

- **Primary use:** Research on explainable automated sleep staging, especially in resource-constrained settings.
- **Intended users:** Sleep medicine researchers, clinical informatics researchers, and AI/ML researchers working on interpretable medical AI.
- **Deployment scenario:** Single consumer-grade GPU inference (e.g., NVIDIA RTX 4090, 24 GB).
- **Clinical note:** This model is intended for research purposes. It has not been validated for clinical diagnostic use and should not replace professional sleep technologist scoring in clinical settings.

## Citation

If you use SleepVLM in your research, please cite:

```bibtex
@article{deng2026sleepvlm,
  author    = {Deng, Guifeng and Wang, Pan and Wang, Jiquan and Li, Tao and Jiang, Haiteng},
  title     = {{SleepVLM}: Explainable and Rule-Grounded Sleep Staging
               via a Vision-Language Model},
  journal   = {}, % TODO: update after publication
  year      = {2026}
}
```

## License

This model is released under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0).