Local Llama — In-Browser AI Chat

Chat

Selected modelonnx-community/Llama-3.2-1B-Instruct-q4f16

Backendauto-detecting…

Quantq4f16

Max new tokens Cap on response length

512

Temperature 0 = greedy · higher = more random

0.70

Top‑p Nucleus sampling cutoff

0.95

Top‑k 0 disables · higher = wider

Repetition penalty 1.0 = none · 1.2+ discourages loops

1.10