omg
Typed questions in, calibrated probabilities out.
Give the model a state and a set of typed questions (choice, score, noul). Models answer without text generation. Everything runs locally through WebGPU, or on the CPU with Bekko; your inputs stay in the browser.
Model
Not loaded. Weights stream from the Hugging Face Hub and are cached by the browser.
Bekko is an experimental English decision model (2–64 options per question). Each question uses its trained task head; mode, calibration and order averaging do not apply. Long inputs are truncated to the model’s token limits. · Original demo
Request