Modelpublic

sdkv2/ministral-3-3b-reasoning-mlx-q8

sourceHugging Faceapache-2.0updated 2mo agoView on Hugging Face

0likes149downloads

Model Card

Ministral-3-3B-Reasoning (MLX, q8)

Q8-quantized MLX variant of sdkv2/ministral-3-3b-reasoning-mlx (itself a conversion of mistralai/Ministral-3-3B-Reasoning-2512, Apache-2.0).

Used for the longer repetition runs in the multiheadtest project where the bf16 6.4 GB model would be too slow / memory-bound on the M4 (16 GB).