4 articles found
Xiaomi quietly dropped MiMo-V2.5-DFlash on Hugging Face. A 311B parameter model with block diffusion speculative decoding that could double your inference speed. Here’s what the community is finding.
Xiaomi’s MiMo V2.5 hits 3000 tps with a 1-trillion-parameter model using a radical FP4 quantization and a ‘block-diffusion’ drafter. Here’s the tech that made it happen and the catch.
Xiaomi’s MiMo-V2.5-Pro doesn’t just crunch code, it outplays humans at complex social manipulation, and you can run it on your own hardware.
An in-depth look at how Xiaomi’s modestly-sized MoE model delivers elite performance at a fraction of the cost, and why the community isn’t buying it.
