Latest ContentInference Cookbook Model Library

MiMo V2 Flash

MiMo-V2-Flash is a large-scale Mixture-of-Experts (MoE) language model developed by Xiaomi for high-speed reasoning, coding, and agentic workflows. The model features 309B total parameters with 15B activated, built on a 48-layer transformer architecture with 64 attention heads and a 4,096 hidden size. It utilizes a hybrid attention design that interleaves Sliding Window Attention and Global Attention in a 5:1 ratio, significantly reducing KV-cache memory while maintaining long-context performance. The model supports up to a 256K token context window and integrates Multi-Token Prediction (MTP) to accelerate generation throughput.

Type	MoE LLM
Capabilities	Text Generation, Instruction Following, Reasoning, Mathematical Reasoning+5 more
Group Release Date	December 15, 2025
Links	Paper\|Blog\|HF Model Card
License	MIT

Inference Instructions

Deploy and run this model on NVIDIA B200 GPUs using the command below. Copy the command to get started with inference.

CONSOLE

docker run -it --rm 
 --runtime=nvidia 
 --gpus all 
 --ipc=host 
 --shm-size=128g 
 -p 8000:8000 
 -v ~/.cache/huggingface:/root/.cache/huggingface 
 -e HF_TOKEN='YOUR_HF_TOKEN' 
 -e LD_LIBRARY_PATH='/usr/local/nvidia/lib64:/usr/local/nvidia/lib:/usr/lib/x86_64-linux-gnu' 
 vllm/vllm-openai:v0.15.0-cu130 
 XiaomiMiMo/MiMo-V2-Flash 
  --tensor-parallel-size 4 
 --max-model-len auto 
  --gpu-memory-utilization 0.95 
 --max-num-batched-tokens 65536 
 --tool-call-parser qwen3_xml 
 --reasoning-parser qwen3 
 --enable-auto-tool-choice 
 --max-num-seqs 1024 
 --disable-log-requests 
 --generation-config vllm 
 --trust-remote-code