license: apache-2.0
language:
- en
library_name: transformers
pipeline_tag: image-text-to-text
base_model: Qwen/Qwen2.5-VL-3B-Instruct
tags:
- sleep-staging
- polysomnography
- PSG
- explainable-AI
- AASM
- vision-language-model
- medical
- EEG
- EOG
- EMG
- lora
- rule-grounded
- multimodal
datasets:
- Feng613/MASS-EX
model-index:
- name: SleepVLM-3B
results:
- task:
type: image-text-to-text
name: Sleep Stage Classification
dataset:
name: MASS-SS1
type: mass-ss1
metrics:
- name: Accuracy
type: accuracy
value: 0.835
- name: Macro-F1
type: macro-f1
value: 0.793
- name: Cohen's Kappa
type: cohens-kappa
value: 0.767
- task:
type: image-text-to-text
name: Sleep Stage Classification
dataset:
name: ZUMS
type: zums
metrics:
- name: Accuracy
type: accuracy
value: 0.812
- name: Macro-F1
type: macro-f1
value: 0.766
- name: Cohen's Kappa
type: cohens-kappa
value: 0.743
SleepVLM-3B
Explainable and Rule-Grounded Sleep Staging via a Vision-Language Model
Paper (coming soon) | GitHub | MASS-EX Dataset | Quantized Version (W4A16) | Collection
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 is a rule-grounded vision-language model for explainable automated sleep staging from polysomnography (PSG) recordings. Unlike conventional black-box classifiers that output only a stage label, SleepVLM generates clinician-readable natural-language rationales citing specific AASM scoring rules for every 30-second epoch, making each staging decision auditable against the clinical standard.
The model takes rendered multi-channel PSG waveform images as input (three consecutive 30-second epochs) and produces a predicted sleep stage (W/N1/N2/N3/R), applicable AASM rule identifiers, and a structured natural-language rationale.
SleepVLM-3B is fine-tuned from Qwen2.5-VL-3B-Instruct through a two-phase training pipeline: waveform-perceptual pre-training (WPT) followed by rule-grounded supervised fine-tuning (SFT) using expert annotations from MASS-EX.
Model Details
| Property | Value |
|---|---|
| Base model | Qwen2.5-VL-3B-Instruct |
| Parameters | ~3.1B |
| Model size | 7.1 GB (BF16) |
| Fine-tuning method | LoRA (r=16, alpha=32, dropout=0.05) |
| Training hardware | 8x NVIDIA A100 80GB |
| Precision | bfloat16 |
| 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 from PSG recordings.
- Intended users: Sleep medicine researchers, clinical informatics researchers, and AI/ML researchers working on interpretable medical AI.
- 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:
@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.