PyTorch’s Helion DSL now targets TPU kernel authoring through Pallas

PyTorch · x · 2026-07-24

PyTorch says Helion, its DSL for custom kernels, is expanding toward hardware-heterogeneous kernel authoring.

• Meta and Google show how Helion compiles to TPU via Pallas.
• The idea is to give developers a unified way to write performant TPU kernels inside the PyTorch ecosystem.
• The blog uses FlashAttention as a case study.
• The compiler autotunes across multiple code-generation strategies, choosing pipelining schemes based on input shapes to maximize compute–memory overlap.

Original post →

More from Infra

• Cloud release velocity scores: AWS 95, Azure 90, Google Cloud 82 — DavidLinthicum · 2026-10-27
• Selling a 5090 to scale local inference: one business weighs 5 hardware paths — Anarchaotic · 2026-10-27
• AWS tutorial: fleet-wide edge CV model lifecycle with SageMaker SDK v3, ~$0.07 per run — AWS ML Blog · 2026-10-27
• Qwen3.8-27B hits 40 tok/s on Jetson Thor with INT4 + MTP, accuracy beats FP16 — No-Dragonfly6246 · 2026-10-27
• Celesto launches CI runners nearly 90% cheaper than GitHub Actions, with agent sandboxes — aniketmaurya · 2026-10-27
• Supabase founders on rebuilding the backend for a billion AI-built apps — Latent Space · 2026-10-27

Infra channel →