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.
Open-source implementation and research project inspired by TypeSafe's Jev and System One. Trains and serves spark-s1 models on DGX Spark, with JevBench evaluation, tool-call safety, and a decision-data factory.
该仓库是一个深入研究 TypeSafe Jev / System One 的开源实现项目,包含模型训练、JevBench 评测、与 Jev 的对比实验、工具调用安全评估及决策数据工厂,实质内容充分。