# PyTorch explains compile-time heuristics that pick Helion kernel configs without autotuning

> Original title: Fast by Default: Compile-Time Heuristics for Helion Kernels

- Source: PyTorch Blog
- Published: 2026-11-12T16:45:26.000Z
- Bad Influence: https://badinfluence.news/items/ga38p50qg9wr0exykcv1ndc68
- Original: https://pytorch.org/blog/fast-by-default-compile-time-heuristics-for-helion-kernels/

## Summary

PyTorch says Helion now uses compile-time heuristics that analyze a kernel's IR and dataflow to choose block sizes, warps, and stages without autotuning. Kernels are classified as matmul, reduction, or pointwise, and a family-specific heuristic emits the config. On H100, the authors report the heuristic reaches 1.18x of hand-written Triton in Flash Linear Attention and 1.27x of vLLM's CUDA reduction kernels. The heuristics are enabled by default on H100 and B200 with the Triton backend.