Program-as-Weights: 0.6B Model Matches Qwen3-32B Prompting With 1/50 Memory
yuntiandeng · x · 2026-09-18
Yuntian Deng (Hugging Face) and collaborators published "Program-as-Weights (PAW)", a new paradigm that compiles fuzzy functions—log alerting, JSON repair, intent ranking—from natural-language specs into compact, locally-executable neural programs.
- A 4B compiler trained on FuzzyBench (a released 10M-example dataset) emits parameter-efficient adapters for a frozen lightweight interpreter.
- A 0.6B Qwen3 interpreter running PAW programs matches direct prompting of Qwen3-32B while using 1/50 the inference memory and running at 30 tokens/s on a MacBook M3.
- Key reframe: the foundation model becomes a tool builder—invoked once per function definition; subsequent calls are cheap, offline, and reproducible.
- Paper, code, and model weights are all public; hosted inference (150 ms remote) is available, with optional local inference for data control.
More from Models
- Hobbyist rewrites Qwen 0.6B weights with 39.8M tokens, no gradients, in 3h12m — Altruistic_Heat_9531 · 2026-11-04
- Same leak, reshared: ChatGPT's five family plan prices spotted again — heyshrutimishra · 2026-11-04
- KOL visits StepFun's Shanghai office, hints a powerful new model is coming — teortaxesTex · 2026-11-04
- Fable 5.5 delayed as Anthropic reportedly bets OpenAI's Astra 6.2 won't beat it — haider1 · 2026-11-04
- ByteDance open-sources Axion 1.0-35B agent foundation model with full reproducible recipe — pmttyji · 2026-11-04
- Community asks Moonshot when Kimi K3.1 model weights will drop — bgmshana · 2026-11-04