openmhc
representation-learning
wearables

OpenMHC Outcome Prediction β€” WBM

Track 1 (outcome prediction) reference checkpoint for the MyHeartCounts / OpenMHC wearable-health benchmark.

This checkpoint is the WBM encoder β€” a bi-directional Mamba2 contrastive self-supervised model that maps a week of wearable sensor data (168 hourly steps, 19 channels) to a 256-d representation. The reported WBM model pairs this encoder (per-user pooled β†’ PCA-50 β†’ linear probe) with a Linear fallback for users without a weekly embedding.

This is an OpenMHC reimplementation of Apple's WBM (Wearable Behavior Model), introduced in Beyond Sensor Data: Foundation Models of Behavioral Data from Wearables Improve Health Predictions (Apple, 2025; see references below).

Pretrained with a contrastive objective on the MHC training split.

  • Checkpoint format: PyTorch Lightning checkpoint (model.ckpt) + normalization_stats.json (canonical hourly z-score constants; channels 0–6 normalized, 7–18 identity).
  • Outcome-prediction tasks: 33 health & behavior labels (classification, ordinal, regression).

Intended use

Intended use: research, methodological comparison, and reproduction of the OpenMHC benchmark.

Out of scope: diagnosis, screening, prognosis, prospective risk prediction, clinical decision support, or any other medical decision-making. OpenMHC's medical condition labels are self-reported and have no diagnosis dates, and benchmark performance does not establish clinical utility. Any diagnostic, prognostic, or screening claim requires independent validation using temporally aligned, clinically adjudicated outcomes in the target population.

Model & implementation

Requirements

Running the encoder needs the CUDA-only Mamba2 kernels (mamba-ssm) and a GPU.

Usage

import openmhc
from openmhc.encoders import WBM

# pip install "openmhc[hf]"  (+ mamba-ssm on a CUDA machine)
enc = WBM.from_release("hf://MyHeartCounts/openmhc-wbm-dp@v1.0")
results = openmhc.evaluate_prediction(enc, version="full")

See openmhc_manifest.json for provenance (source W&B artifact, training details) and architecture metadata.

License

Released under the BigScience OpenRAIL-M license. The use restrictions in its Attachment A apply to this checkpoint and any derivatives; among other things, they prohibit providing medical advice and medical results interpretation.

Citation

If you use this checkpoint, please cite the OpenMHC benchmark.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Space using MyHeartCounts/openmhc-wbm-dp 1

Paper for MyHeartCounts/openmhc-wbm-dp