vLLM adds day-0 support for Mistral Large 4 with DSpark and NVFP4
AIvLLM announces day-0 support for Mistral Large 4, a 1-trillion-parameter mixture-of-experts multimodal model with 49 billion active parameters per token. The post reports about 2x output throughput on TP8 with the built-in DSpark drafter and about 1.5x more KV cache with the NVFP4 checkpoint, which stays within 2.1 points of FP8 on evaluated tasks.
AIWhy it matters.
The post gives specific architecture, memory, and throughput figures for running Mistral Large 4 on vLLM, useful for judging serving tradeoffs and deployment layouts.
Loading source posts…
Source details
vLLM Blog · Mistral and the vLLM Team · vllm.aiPublished · added here
Original title: Day 0 support for Mistral Large 4 on vLLM