Stanford and NVIDIA's CLM embeds decisions instead of tokens, 13x faster but accuracy drops at scale
Prompt Engineering · youtube · 2026-09-26
CLM: embedding-match instead of token generation
Stanford and NVIDIA Research introduced Contrastive LM (CLM), a "System 1" model that works like CLIP for decisions: it embeds the situation and every candidate action into one space and picks the nearest match instead of generating tokens, making decisions 13x faster.
Training and benchmarks:
- Built on a frozen 8B backbone with only small heads being trained
- Tested on a DGX Spark across 1,080 tools: latency held around 80ms, but accuracy fell from 86% with 8 tools to 17% with all of them
Positioning: not a replacement for reasoning models today, but a fast first-stage shortlist before a model that compares options side by side.
Code, weights (8B) and a playground are open-sourced.
More from Infra
- Running Qwen2.5-Coder-3B on Android: Jupyter + Pydantic Evals on-device eval pipeline — alex_b3b · 2026-11-11
- 22% of Nvidia revenue now books to Singapore, but only ~1% is local buyers — Gregory_C_Allen · 2026-11-11
- Huawei's Kirin 9050 NPU claims offline 30B MoE on smartphones, +140% perf — AIFlow_ML · 2026-11-11
- Stratechery: Intel vs AMD's diverging earnings and Intel's need for an old-new path — Stratechery · 2026-11-11
- a16z Podcast: Mike Volpi on What the 1990s Internet Boom Teaches Us About the AI Buildout — a16z Podcast · 2026-11-11
- Alibaba files to list chip-design unit T-Head in Hong Kong to fund AI push — pstAsiatech · 2026-11-11