typevet¶

Kind: reference, the landing page. Choose a path by what you need to do.

typevet is a Python library that asks a model typed questions and returns typed answers. The three question types are Noul (yes or no), Choice (one label) and Score (one rubric level). typevet computes each answer from the model's next-token probabilities, read before sampling. typevet also returns JSON objects that pass a JSON Schema you supply, or it raises an error. The receipts cover Gemma 4 31B on llama.cpp for local work and on vLLM for hosting.

typevet is pre-1.0. Install it from PyPI with pip install typevet. The full index lists every page and its kind.

Start with the offline tutorial

Choose a path¶

Run offline, with no model

Install typevet and get typed answers from a scripted fake.
Evaluate typevet

Read what typevet does, what each receipt proves and where the limits are.
Integrate typevet from Python

Find the public surface, the settings and the errors.
Host typevet on vLLM

Serve the tested vLLM pin on one H100, your own or rented.
- Serve typevet on vLLM
- Serve Gemma 4 31B on a rented H100
Run typevet on llama.cpp locally

Run a stock llama-server with nested json_schema and opt-in live tests.
- Run Gemma 4 on llama.cpp
Send images with a question

Send an image to each backend and check on llama.cpp that it changes the answer.

Contribute¶

Testing pyramid: the unit, contract and live layers.
Writing system: page kinds and prose rules.
Commit messages: the Conventional Commits vocabulary.
Contributor and agent notes: the gates, the issue workflow and the non-negotiable rules.