Jevals Data: Independent benchmark for TypeSafe Jev
Independent benchmark data comparing TypeSafe AI's Jev (System One model) against LLMs on decision score, accuracy, calibration, cost, and latency for typed tasks.
A learning project that implements a TypeSafe Jev/System One-style decision endpoint on a Qwen3-0.6B backbone with LoRA, plus an OpenAI-compatible chat/generation server and MiniJev-compatible evaluation. Not production software.
README 明确以 TypeSafe Jev/System One 的概念和 API 规范为基础,实现了 /v1/systemone 决策接口,并在 Qwen3-0.6B + LoRA 上进行训练、服务与评测,包含 MiniJev 对照和 holdout 结果,属于真实深入的研究/实验项目。