gemma-4-e4b-glucolens-rollout

LoRA adapter on top of google/gemma-4-e4b-it that returns a strict JSON 6-hour glucose projection (12 trajectory points, 3 ranked drivers, interpretation, history grounding, overall confidence) for a given meal and CGM context.

Submission for The Gemma 4 Good Hackathon, Health track, May 2026.

Full documentation, training procedure, eval methodology, schema, demo, and limitations: https://github.com/kotlarmilos/gemma4-glucolens

Quick start

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

BASE    = "google/gemma-4-e4b-it"
ADAPTER = "kotlarmilos/gemma-4-e4b-glucolens-rollout"

tok   = AutoTokenizer.from_pretrained(BASE)
base  = AutoModelForCausalLM.from_pretrained(BASE, torch_dtype=torch.bfloat16, device_map="auto")
model = PeftModel.from_pretrained(base, ADAPTER)

See demo.ipynb for the prompt template and five worked scenarios.

Headline numbers

n=50 in-distribution test + 50 OOD, greedy decode. Full table and methodology in the repo.

Metric	Base zero-shot	Fine-tuned
Trajectory MAE (mg/dL)	19.9	6.9
Peak timing error (min)	6.6	1.2
OOD refusal rate	0.00	1.00

Two known failures (calibration ECE 0.30, counterfactual consistency 0.00) are documented in the repo README, Section 6.

Training data

Synthetic only, ~15k rollouts from kotlarmilos/glucolens-rollout-triples. Generated by src/sim.py (simglucose + Bergman ODE fallback). No real patient data.

Not a medical device

Research code. No diagnosis, treatment, or insulin guidance. Aimed at non-diabetic and pre-diabetic metabolic awareness only.

License

Apache 2.0 for adapter weights and config. Base model governed by the Gemma usage policy.

Citation

@misc{kotlar2026glucolens,
  author = {Milos Kotlar},
  title  = {GlucoLens: 6-hour glucose rollouts with a fine-tuned Gemma 4 E4B},
  year   = {2026},
  howpublished = {The Gemma 4 Good Hackathon},
  url    = {https://github.com/kotlarmilos/gemma4-glucolens}
}