1. Install
  3. 2 Example
  5. 3 Your data
Ready in 60 seconds

Install the
Standard Model.

Two ways to get the open-weight JEPA model for longitudinal EHR running on your machine. Pick the quickstart for an end-to-end demo, or pull weights directly from Hugging Face.

Two paths

Quickstart, or straight from the source.

Recommended

~60 s · end-to-end

Quickstart

A few lines install uv, clone the Quickstart repo, sync dependencies, and run the MIMIC-IV demo on 100 patients.

terminal 
$ curl -LsSf https://astral.sh/uv/install.sh | sh
$ git clone https://github.com/standardmodelbio/quickstart.git
$ cd quickstart && uv sync
$ uv run python demo.py
  • Installs uv if missing
  • Clones the quickstart repo
  • Pulls smb-v1-1.7b from HF
  • Runs the 4-task linear probe demo
See the full example

Bring your own stack

Weights only

Hugging Face

Load the model directly with transformers. No demo, no opinions about your environment.

python 
# pip install transformers torch
from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained(
    "standardmodelbio/smb-v1-1.7b",
    trust_remote_code=True,
    device_map="auto",
)
  • All 5 SMB-v1 variants on the same org page
  • Bring your own MEDS pipeline + serialization
  • Works with conda, pip, poetry, anything
Browse on Hugging Face
What you'll see

From curl to clinical predictions.

The quickstart embeds 100 MIMIC-IV demo patients and reports linear-probe scores for four canonical clinical tasks. Below is the live terminal output you'll see.

cd quickstart && uv run python demo.py
[1/4] Loading MIMIC-IV demo data
      916,166 events · 100 subjects
[2/4] Loading smb-v1-1.7b
      model.safetensors  3.66 GB
[3/4] Generating embeddings
      100/100 patients · last-token pooling · dim=2048
[4/4] Clinical task heads
      Readmission    ROC-AUC    0.842
      Phenotype      Accuracy   0.761
      Survival       MAE        4.18 months
      Cox PH         C-Index    0.713
  1. 1

    MEDS events

    916k timestamped clinical events across 100 patients.

  2. 2

    Serialize

    smb_utils converts events into causal token streams.

  3. 3

    Embed

    smb-v1 encodes each patient into a 2048-dim vector.

  4. 4

    Predict

    Linear probes trained on four clinical outcomes.

Reference

Going further.

Manual installation, verifying the install, memory & quantization, and common troubleshooting.

Manual installation

Requirements: Python 3.11+, Git.

bash 
curl -LsSf https://astral.sh/uv/install.sh | sh
git clone https://github.com/standardmodelbio/quickstart.git
cd quickstart && uv sync
uv run python demo.py

Prefer conda or pip? See pyproject.toml in the Quickstart repo.

Verify the install

A quick sanity check that CUDA is visible and the weights load.

python 
uv run python -c "
import torch
from transformers import AutoModelForCausalLM

print(f'PyTorch: {torch.__version__}')
print(f'CUDA:    {torch.cuda.is_available()}')

model = AutoModelForCausalLM.from_pretrained(
    'standardmodelbio/smb-v1-1.7b',
    trust_remote_code=True,
    device_map='auto',
)
print('Model loaded.')
"
GPU memory & quantization

Match the precision to your hardware. 8-bit and 4-bit require bitsandbytes.

PrecisionVRAMGPUCode
float3216 GBA100default
float168 GBRTX 4090torch_dtype=torch.float16
8-bit4 GBT4load_in_8bit=True
4-bit2 GBRTX 3060load_in_4bit=True
Troubleshooting
CUDA not detected
Run nvidia-smi. Update NVIDIA drivers if needed.
Out of memory
Use float16 or quantization from the table above.
Access denied
Run huggingface-cli login with a valid token.
Slow download
Weights are ~3.7 GB. Ensure a stable connection.