vLLM Trends at 79,633 Stars as Dominant Open-Source LLM Inference Engine — Model Runner V2 Rewrite and Blackwell GPU Support Drive Continued Adoption

GitHub·medium signal

vLLM maintains its position as the go-to open-source inference engine with 79,633 stars, featuring PagedAttention and continuous batching for high-throughput memory-efficient LLM serving. The March 2026 Model Runner V2 ground-up rewrite delivered significant performance improvements, with the engine now supporting NVIDIA Blackwell GPUs, DeepSeek V3, and AMD ROCm acceleration. An independent May 2026 inference engine benchmark comparing vLLM ROCm, TurboQuant, and LLaMA.cpp confirms vLLM's throughput leadership on supported hardware.