PyTorch Helion Kernels Boost vLLM Inference Throughput

PyTorch, Meta, Red Hat, and the vLLM team integrated the Helion kernel DSL as vLLM's linear-layer backend on NVIDIA Hopper, boosting throughput by over 10% on some workloads and speeding up kernels by 1.11-1.18x.

2026-10-03 ~ 2026-10-03 · 2 related posts

• PyTorch Helion DSL integrated into vLLM beats CUTLASS and DeepGEMM with 10%+ throughput gains — PyTorch · 2026-10-03
• Helion-powered vLLM linear backend on Hopper yields 1.11-1.18x kernel speedups — lmoroney · 2026-10-03