Meta-Planner WebGPU Client

This page runs inference in the visitor browser using transformers.js + WebGPU. Meta-Planner currently uses OpenForecaster-8B and needs an ONNX/WebGPU-compatible repo to run fully in-browser.

Model repo (ONNX/WebGPU compatible)

Quant option

Max new tokens

Quant mapping in browser mode: IQ2_XXS→q4, IQ3_XXS→q8, IQ4_XS→q4f16, BF16→fp16.

Message

Ready.

Auto-save confident predictions Confidence threshold (0.70)

🧠 Meta-Planner Browser WebGPU Client

Saved Bets (browser local)