Local experiment · Third-party
jevlike
Independent starter model with a Jev-like input/output shape: text plus options in, probabilities out. Not TypeSafe Jev.
- Runnable
- Runnable locally with Python and uv. Text training does not need a TypeSafe key. Some vision demos expect Apple Silicon / MPS.
- Keys
- No TYPESAFE_API_KEY. Optional game extras and local checkpoints only.
- Platforms
- Python, local training, Apple Silicon for some vision demos
- Primitives
- None — local Jev-like experiment, not TypeSafe primitives
uv pip install -e '.[dev]'If you want to study the shape of a decision model without calling a hosted API, jevlike is the honest local object: a piece of text, N options, one pass, a probability per option.
The repository includes synthetic data commands, training, evaluation, and predict CLIs. Vision examples reuse an option-attention head over image patches.
The README states TypeSafe has not published Jev's design. Keep jevlike labeled as a local experiment with a similar type signature, not as hosted Jev.
Caveats
- The README states TypeSafe has not published Jev's design. This is an independent architecture with a similar type signature.
- Do not present jevlike outputs as official Jev answers.
- Vision and game examples are research demos, not production classifiers.