Home▸AI & Machine Learning▸3D Function-Call Arguments: Constrained Decoding Masks, Rejection Retries with Hints and the KL Cost

🧩 3D Function-Call Arguments: Constrained Decoding Masks, Rejection Retries with Hints and the KL Cost

Four rows of probability bars on a Blender-built workbench, one per function-call argument: green bars are schema-valid, red bars are not. Compare free sampling, rejection with retry hints and constrained decoding, and see that the mask costs exactly log2(1/P) bits of KL, with Monte Carlo, geometric-series and identity checks.

AI & Machine Learning3DModerate60 FPS📱 Mobile-adapted🟠 Blender⇄ 2D version
3D Function-Call Arguments: Constrained Decoding Masks, Rejection Retries with Hints and the KL Cost ↗ Open standalone
⚙ Under the hood

Four rows of probability bars on a Blender-built workbench, one per function-call argument: green bars are schema-valid, red bars are not. Compare free sampling, rejection with retry hints and constrained decoding, and see that the mask costs exactly log2(1/P) bits of KL, with Monte Carlo, geometric-series and identity checks.

Constrained DecodingFunction CallingJSON SchemaRejection SamplingToken MaskingKL DivergenceLarge Language ModelsRetry Budget

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install

What did you find?

Add reproduction steps (optional)