whereismyai
SemIf OpenJev
SemIf OpenJev is an open-source research tool that evaluates semantic decisions directly from open models using typed option probabilities.
SemIf OpenJev is an independent research project that reproduces runtime-defined semantic decision interfaces using open models. It reads typed option probabilities directly from a model forward pass instead of sampling answer tokens or parsing text. The tool supports multiple backends including Python with CUDA, Apple Silicon via MLX or PyTorch/MPS, and a llama.cpp CPU backend with GGUF checkpoints. It includes features for shared-state prefilling, per-workload temperature calibration, and browser-based demos. It is built for developers and researchers who need auditable, decision-native outputs from language models without generation overhead.
SemIf OpenJev at a glance
- Best for: Software engineer, Researcher
- Runs on: Web, macOS, Windows, Linux, API, CLI, self-hosted, open source
- Pricing: Free under the MIT License.
- Released: September 2026
- Category: AI research tool
- Website: github.com/theoleecj/semif-openjev
Key features
- Direct typed logit scoring without text generation
- Shared-state prefilling for parallel criterion evaluation
- Apple Silicon support via MLX and PyTorch/MPS
- CPU scoring with GGUF checkpoints via llama.cpp
- Per-workload temperature calibration
- Browser-based WebGPU demo
- Auditable fixtures and row-level outputs
Use it for
- Routing agent decisions
- Evaluating conditional logic statements
- Running semantic benchmarks on local hardware
- Calibrating model decision confidence
Alternatives to SemIf OpenJev
- NanoJev — It runs parallel decisions and outputs probability distributions without output-token decoding.
- Jev — Fast, cheap typed decisions that software can act on directly
- Jiwo — It provides low-latency decision-making with a 50ms median request time.
- Decido — Runs locally on CPU while beating published benchmark scores.
- Kev — Provides calibrated probabilities and matches the TypeSafe API.
- Vev — Runs locally on NVIDIA hardware with a fast, Jev-compatible API.
- Laya — It runs typed decisions in a single forward pass over 100+ languages.
- Project STROT — Runs locally at zero marginal cost with typed JSON outputs and confidence-based abstention.
- OpenResearch — Run parallel research workflows locally or on remote compute.
- Clef — Offers low latency and high-speed classification for agentic workflows.
- OpenCompass — It offers an objective, comprehensive, and credible open-source AI evaluation system.
- Vegaml — Runs locally on your hardware with calibrated probabilities and conformal answer sets.
All SemIf OpenJev alternatives