Xiaomi Unveils MiMo-V2.5-DFlash Model
·2026.07.12 16:11
Key point
Xiaomi has released the MiMo-V2.5-DFlash model weights, which have a large number of parameters, via Hugging Face.
Details
Xiaomi has released the MiMo-V2.5-DFlash model weights on Hugging Face. This model is a large-scale model with over 300B (300 billion) parameters.
Key features are as follows:
- Model weights are provided through the DFlash directory.
- It includes an MTP (Multi-Token Prediction) head and a separate MTP model.
- It has a structure that can improve or supplement the previous structure that didn't work in Llama.cpp and others due to MTP layer recognition issues.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.