README.md

metadata

model-index:
  - name: Gluformer-tiny
    results:
      - task:
          type: glucose-prediction
        metrics:
          - name: RMSE
            type: 60 minute prediction
            value: 25.36
        source:
          name: Brown2019
          url: https://www.nejm.org/doi/full/10.1056/NEJMoa1907863

Model Card for Gluformer Blood Glucose Prediction Model

This model uses past continuous glucose monitor (CGM) values to predict values for the next hour.

Model Details

Model Description

• Developed by: Renat Sergazinov, Mohammadreza Armandpour, Irina Gaynanova
• Funded by: Texas A&M University
• Shared by: Nat Jeffries
• Model type: Time series encoder-decoder Transformer

Model Sources

• Repository: Github
• Paper: Arxiv

How to Get Started with the Model

Use the code below to get started with the model.

from transformers import AutoModel, AutoConfig
from datetime import timedelta, datetime

model = AutoModel.from_pretrained('njeffrie/Gluformer-tiny', trust_remote_code=True)
config = AutoConfig.from_pretrained('njeffrie/Gluformer-tiny', trust_remote_code=True)

# Dummy input and timestamp values.
input_glucose = [100.0 for _ in range(config.len_seq)]
timestamps = [datetime(2025, 7, 25, 0, 0) + timedelta(minutes=5 * i) for i in range(len(input_glucose))]
subject_id = 0

pred, log_var = model(subject_id, timestamps, input_glucose)

Predictions will be predicted future glucose values in 5 minute increments. Log var indicates confidence. See the paper for more details.