2026-11-06 · America/Los_Angeles · 社区动态 · #11

AMA: How we trained a SOTA Decision model, got top 3 on OpenRouter & top on 5 ArtificalAnalysis: full playbook - arch, datasets, training, evals, inference

hey llamas! My name is Alt, and I'm on the technical staff at Nace AI . We're doing a quick AMA here, so ask us anything and we'll answer. we just published a full technical report on how we build, train, and evaluate Drex, our family of four open decision models. A decision model takes a state (a document, record, or agent context, optionally with images) plus questions with options, and returns a probability distribution over the options in a single pass, with no autoregressive decoding and no output parsing. Top 3 on OpenRouter Top 5 on Artificial Analysis 1. Models - Drex DLM (8B):…

阅读原文查看评分、趋势与关联来源阅读整期日报

AMA: How we trained a SOTA Decision model, got top 3 on OpenRouter & top on 5 ArtificalAnalysis: full playbook - arch, datasets, training, evals, inference

热度 47.4 / 100；排名与评分保留该期记录。