Skip to content
Other

DiabAI 30-Minute Glucose LSTM

30-minute blood glucose forecasting

2 min readupdated

A compact, educational PyTorch model that predicts blood glucose levels 30 minutes ahead using continuous glucose monitoring (CGM) data. It takes 96 consecutive readings from the past 24 hours (at 15-minute intervals), the current time of day, and the diabetes type (T1DM or T2DM), outputting the projected change in mg/dL. Under the hood is a single-layer LSTM (a recurrent architecture built for time-series data) with just 19,617 trainable parameters. Trained on the open ShanghaiT1DM and ShanghaiT2DM datasets, it is aimed at researchers, students, and developers building non-clinical prototypes to experiment with medical time series.

Key features

  • Delta prediction: Instead of predicting the absolute final value, the network estimates the 30-minute glucose change and adds it to the latest reading.
  • Circadian rhythm awareness: Alongside two time-series channels (normalized glucose and its first difference), the model takes the sine and cosine of the hour of day plus a type 1 diabetes indicator.
  • Beats classical baselines: On a held-out test set, the LSTM achieved a mean absolute error (MAE) of 8.7502 mg/dL, outperforming linear regression (9.2644) and random forest (8.8821).
  • Ultra-lightweight: With fewer than 20,000 parameters, it runs easily on a standard CPU without a GPU.

What you can do

  • Retrospective data analysis: Test forecasting behavior on historical CGM sensor logs.
  • ML teaching sandboxes: Use the repository as a reproducible benchmark to compare RNNs, CNNs, and transformers on time-series tasks.
  • Non-clinical prototypes: Plug the inference script into research dashboards to visualize glucose trends (strictly with human oversight and no medical advice).

Example

import json
from inference import load_model, predict_future_glucose

model, normalization = load_model("model.pt")
document = json.load(open("example_input.json", encoding="utf-8"))
result = predict_future_glucose(model, normalization, document)
print(result)
Read moreDetails
Test metrics (16,992 samples, 17 patients): MAE = 8.7502 mg/dL, RMSE = 12.6120 mg/dL, R² = 0.9335 (a 23.25% MAE reduction over a naive forecast).
Accuracy by range: Performs best within 70–180 mg/dL (MAE 7.9208) and drops above 180 mg/dL (MAE 13.0000).
Hardware: Trained and runs on CPU (PyTorch 2.15.0, Python 3.11.9); requires a complete 24-hour sequence of 96 data points with no missing values.
Limitations: Not a medical device; must not be used for diagnosis, treatment, or insulin dosing; does not account for meals, physical activity, or insulin injections; CC BY 4.0 license.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # noracodes/diabai-glucose-forecasting-lstm
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download noracodes/diabai-glucose-forecasting-lstm --local-dir ~/models/diabai-glucose-forecasting-lstm
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of diabai-glucose-forecasting-lstm (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

noracodes/diabai-glucose-forecasting-lstm
Author
noracodes
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Other

GyroFlow

GyroFlow