标签：huggingface

Open reproduction of TypeSafe Jev: a 150M typed decision engine (noul/choice/score in one non-autoregressive pass, calibrated confidence). 0.697 vs Jev's 0.727, 2.5x better calibrated, 4x faster, free. Trains on a Colab T4 in 30 min.

A decision model that returns calibrated answers to yes/no, choice, score and multi-label questions.

Single-forward-pass semantic routing in Rust: turn dense LLMs (Llama, Qwen, Mistral, Gemma) into a typed decision API with LoRA/QLoRA training and FP8/FP4 quantization, built on Candle.

Typed decisions from one encoder pass. Open weights, Apache 2.0.

Open 4B typed-decision model (Jev-style /v1/systemone): calibrated per-option probabilities from one prefill. Training, data generation and eval pipeline on HF Jobs.