GPTQ rounding inside existing MLX 4-bit layouts: same size, speed and format as the template, closer to bf16.
-
mlx-community/Qwen3.8-27B-GPTQ-4bit
Image-Text-to-Text • 27B • Updated • 891 • 2 -
sady4850/Qwen3.8-27B-4bit-GPTQ-agentmix
Image-Text-to-Text • 27B • Updated • 115 -
sady4850/gemma-4-E4B-it-MLX-4bit-GPTQ
Image-Text-to-Text • 8B • Updated • 126 -
sady4850/Qwen3.8-27B-4bit-GPTQ-agent
Image-Text-to-Text • 27B • Updated • 126