The Agent Architect · 2026-W47
The Agent Architect #47: Abductive Reasoning
预览:本期将于周二08:00 UTC发送。
收听最新一期 · 7 min
本周模式
Abductive Reasoning
- 是什么:
- 从不完整的数据中生成相互竞争的假设,再根据现有证据按可能性排序。
- 何时使用:
- 在信息不完整时诊断根因、排查故障,或从可观测的现象推断隐藏状态。
- 注意:
- 它偏向最连贯的叙事,而非最可能的解释;确认偏误会让你早早锁死在错误的假设上。
本周智能体AI动态
- DeepSeek-V4.1-Flash on Vera Rubin NVL72: 58x Throughput vs. GB300 NVL72 at 387 tokens/svLLM
DeepSeek-V4.1-Flash achieves 1.7M tokens per second per GPU on one NVIDIA Vera Rubin NVL72 rack at 387 tokens per second P90.
- A local language model layered with rule-based and neural detectors for clinical text de-identification: an open-source toolkit and evaluationarXiv cs.AI
DeidSharp combines local LLM with rule-based and neural detectors to identify protected health information in clinical text.
- When One GPU Isn’t Enough: Serving GLM-5.3 Across Two NVIDIA DGX Stations with vLLMNVIDIA Developer
GLM-5.3 model served across two NVIDIA DGX Stations connected via ConnectX-8 RDMA using vLLM for distributed inference.
- ArchSpace: Recording, Reviewing, and Referencing LLM Architecture ResearcharXiv cs.AI
ArchSpace documents LLM architecture research decisions, tracking which innovations succeeded, failed, or were abandoned across industry and academia.
- Inkling 2 Small on vLLM: Day-0 SupportvLLM
vLLM supports Inkling 2 Small with attention residuals, new decode kernel, fused all-reduce, and DSpark speculative decoding.
The Agent Architect
每周一个模式、一个权衡、一个生产事故案例。为构建智能体系统的人准备的每周简报。
每周一封邮件,一键退订。您的地址仅用于发送简报。