GitHub · Research and data
TheoLeeCJ/SemIf
The open-model route with the measurements attached.
What it does
Typed decisions from open models, run locally — "Semantic ifs from open models, on a 3090 at home". The repository was formerly named OpenJev, says plainly that it is independent and not affiliated with TypeSafe, and packages a measured replay: the same frozen 4B model, the same state and the same 21 questions, with typed decisions appearing together while JSON streams token by token.
The repository's own description: “Semantic ifs from open models, on a 3090 at home. Independent; not affiliated with Jev or TypeSafe.”
How it works
`src/semif_phase1/` holds the implementation; `benchmarks/`, `results/` and `demo/` carry the measurements; `webgpu-demo/` runs a version in the browser with no waitlist and no key.
The repository, by the numbers
Read from the GitHub API on 2026-09-22. Stars and the last push move daily — quote them with the date, the way we do.
What we checked
- the tree: benchmarks, results, demo, webgpu-demo, src/semif_phase1, tests
- MIT LICENSE (TheoLeeCJ)
- README names the project as formerly OpenJev and disclaims affiliation
- README describes the aligned replay of one frozen 4B model over 21 questions
What we did not check. We read the benchmark's naming and framing, not its methodology; nothing here was re-run.