vLLM Ships Built-in DSpark and Mistral Large 4 Quantized Weights with Up to 2.23x Throughput Gains

vLLM rolled out built-in DSpark support, boosting output throughput 1.92–2.23x depending on quantization, and released FP8/NVFP4 quantized weights for Mistral Large 4 (1T-parameter MoE), roughly doubling FP8 throughput.

2026-10-29 ~ 2026-10-29 · 2 related posts

• vLLM's built-in DSpark delivers up to 2.23x output throughput gain — vllm_project · 2026-10-29
• Mistral Large 4 quantized weights hit vLLM with ~2x throughput via built-in DSpark — vllm_project · 2026-10-29