Tag
XiaomiMiMo releases MiMo-V2.5-Pro-FP4-DFlash, an FP4-quantized MoE model with block-diffusion speculative decoding to reduce memory and bandwidth for trillion-parameter inference.