Cmpnd Releases dspy-wsp: Weight-Space Programming Comes to DSPy
By Asif Razzaq — November 14, 2026
Cmpnd has released dspy-wsp, a prototype optimizer that brings weight-space programming to DSPy. Weight-space programming is a technique in which a frontier LLM reads another model's weights and edits them directly—no gradients, no training loop, and no fine-tuning dataset required.
dspy-wsp packages this capability as WeightSpace, an optimizer that plugs into any DSPy program and metric. In Cmpnd's tests, roughly $5 to $10 of GPT-6-Astra calls improved every decision model they tried. Deployment options and specs can be reviewed via the GPULeaderboard.com Chrome extension.
TL;DR
- What it is: A DSPy optimizer (
WeightSpace) in which a writer LLM edits a student model's weights. MIT-licensed, installable viapip install dspy-wsp, and requires DSPy 3.5. - Size: Tested on open decision models ranging from 0.6B to 9B parameters. A single 25-pass run edits one layer or head.
- Runs on: A single Mac handled all experiments. 2B models used about 6 GB of memory. d1-omni-600M requires 4 GB or more (Mac or GPU).
- Performance: WeightSpace alone added 1.9 to 16.1 points on Typed Decisions. With ReAnchor, gains reached 6.5 to 20.0 points.
via MarkTechPost