SLEEPJEV
Jev-inspired PyTorch research framework that encodes an overnight sleep recording once and answers many Choice, Noul and Score questions.
Overview
SLEEPJEV (Sleep-JEV) is an open-source PyTorch research framework for polysomnography (PSG), sleep staging and runtime sleep analysis from PAI-CUHK. It applies Jev's design as a lens: encode the overnight signal once into a reusable representation, build serving indexes over it, read a query-conditioned subset of time tokens, and score the options each runtime question supplies. It supports Choice, Noul and Score primitives, so many explicit sleep questions can be asked of one recording with inspectable probabilities and reuse of the same encoded state across a multi-task workload.
An interactive demo replays verified inference on SHHS feature caches, showing parallel questions, probabilities, confidence, action gates and latency, and a local server runs real checkpoint inference. It is an independent implementation with local weights; it does not call TypeSafe's API. A companion project, MEDJEV, applies the same idea to clinical text. The README is explicit that this is a research prototype, not a medical device or source of clinical advice.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which sleep stage or event category applies in this window?(inferred) | — |
| Yes / no | Did this event occur during the night?(inferred) | — |
| Score | How severe is this sleep finding?(inferred) | — |
How it works
- Encode the overnight PSG signal once
- Index the shared representation for retrieval
- For each runtime question, retrieve relevant time tokens
- Score the supplied options and return typed probabilities
Strengths
- One encoding serves many questions
- Options supplied at query time rather than fixed heads
- Interactive demo and local inference server
Caveats
- Research prototype; not for clinical use
- Independent of TypeSafe's Jev
At a glance
- Type
- Service
- Status
- research
- Language
- Python
- Runtime
- PyTorch research framework + local server
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
JEV-inspired typed runtime decisions over long-horizon PSG: reusable overnight…
wh000wh000/awesome-jev-live
Frequently asked questions
Does SLEEPJEV call TypeSafe's Jev?
No; it is an independent Jev-inspired implementation with local weights.
Is it a medical device?
No; the README says it is a research prototype not validated for patient care.
What is the main idea?
Encode an overnight recording once and answer many typed questions over it at runtime.
Related projects
Rust research workbench where Jev gives a cancer-of-unknown-primary origin distribution over 22 classes, gated locally for review.
ProjectresearchRustOpen clinical text classifiers (0.8B-9B on Qwen3.5) with labels set at inference time, run locally and compared with Jev 1.13.
ProjectresearchPythonExperiment screening ADHD review abstracts as include or exclude with Jev, scored against Cohen et al. 2006 abstract triage labels.
AppresearchPythonTwo frozen open models answer yes/no per exam option; their agreement drives an auto-release gate and a conformal candidate set.
ServiceresearchPythonResearch framework for biomedical decision models on Qwen3.5, studying domain pretraining, NLI training, calibration and abstention.
BenchmarkresearchPythonNextflow plugin exposing Jev noul, choice and score judgments as pipeline functions you can gate on with ordinary operators.
PluginbetaGroovy