Skip to content

Add jevos - #346

Closed
feder-cr wants to merge 1 commit into
mjhea0:mainfrom
feder-cr:add-jevos-1b
Closed

feder-cr wants to merge 1 commit into
mjhea0:mainfrom
feder-cr:add-jevos-1b

Conversation

@feder-cr

@feder-cr feder-cr commented Sep 27, 2026 •

Copy link
Copy Markdown

Project Information

  1. Project Name: jevos

  2. Project URL: https://github.com/feder-cr/jev (release: https://github.com/feder-cr/jev/releases/tag/jevos)

  3. Description:

    jevos is an open-source, self-hosted alternative to TypeSafe's Jev for yes/no decisions. It's a FastAPI server exposing a Jev-compatible HTTP API (/v1/systemone): send a text and a yes/no question, get back P(yes) from a single forward pass of a 1B model (MiniCPM5-1B cut to 17 layers, GGUF q4_k_m, 619MB) running on llama.cpp, CPU-only. TypeSafe's official SDK works unchanged for yes/no questions. On a laptop CPU it answers in 54ms (short request) to 220ms (long request).


Criteria

  1. Is the project new?

    • Yes
  2. How long has the project been maintained?

    The GitHub repo (feder-cr/jev) was created in August 2024, but honestly the current jevos model and server were released 2 days ago (2026-09-25) as a fresh, focused project rather than a year of iteration on this exact artifact. Submitting anyway per this list's openness to new, useful FastAPI projects — happy to have it deferred if a longer track record is preferred.

  3. How many releases has it had if it's a library or package?

    One release so far: jevos (https://github.com/feder-cr/jev/releases/tag/jevos), containing the GGUF model files and server.

  4. Are you the author, or are you submitting the project on behalf of a company?

    • I am the author
  5. What makes it awesome?

    It's a drop-in, open-source, self-hostable replacement for a commercial "yes/no decision" API, using an off-the-shelf FastAPI server + llama.cpp, running entirely on CPU (no GPU needed) with sub-250ms latency. Useful for anyone using TypeSafe's Jev SDK who wants a free/local/open alternative for the yes/no-question use case.


AI Disclosure

  • Yes, this pull request was generated or assisted by AI.

Additional Information

Disclosure: I'm one of the project's two authors (feder-cr). Measured accuracy on 2,000 held-out yes/no rule questions: jevos 0.815 vs. Jev 0.927 vs. Laya 0.489 — lower than the commercial original but functional and free/local. MIT licensed.

🤖 Generated with Claude Code

@feder-cr feder-cr changed the title Add jevos-1b Add jevos Sep 27, 2026
@feder-cr feder-cr closed this by deleting the head repository Oct 3, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant