Official OsaurusAI MXFP8 microscaling (8.25 bpw) build of WeiboAI/VibeThinker-3B (MIT) — a 3B dense reasoning model (Qwen2). Quantized by Osaurus; runs on Apple Silicon via Osaurus / mlx_lm. Emits <think> reasoning blocks.

Usage

python -m mlx_lm generate --model OsaurusAI/VibeThinker-3B-MXFP8 --prompt "What is 17 + 28?"

Or load in Osaurus.

Provenance

Quantization: Osaurus · MXFP8 microscaling (8.25 bpw) · eric@osaurus.ai