NangWeiLun/MiMo-VL-7B-RL-2508-bnb-4bit-fp4

MiMo-VL-7B-RL-2508 — 4-bit BitsAndBytes Quantized

This is a 4-bit quantized version of XiaomiMiMo/MiMo-VL-7B-RL-2508, using the BitsAndBytes library.

Quantization reduces memory usage and makes it possible to run this model on consumer GPUs (≤ 12 GB VRAM), at the cost of a small reduction in generation quality.

Quantization Details

• —Method: BitsAndBytes (bnb)
• —Precision: 4-bit (fp4)
• —Compute dtype: bfloat16
• —Double quantization: disabled
• —Format: safetensors