---
title: "TensorLean formally verifies GPU kernels, finds 38 bugs"
url: https://www.parallelquant.com/posts/tensorlean-formally-verifies-gpu-kernels-finds-38-bugs-f452b8
source_name: "MarkTechPost"
source_url: https://www.marktechpost.com/2026/11/21/columbia-certik-tensorlean-formal-verification-gpu-kernels/
published: 2026-11-22T01:51:43.000Z
topics: ["research", "chips"]
publisher: "Parallel Quant"
---

# TensorLean formally verifies GPU kernels, finds 38 bugs

*2026-11-22 · Source: [MarkTechPost](https://www.marktechpost.com/2026/11/21/columbia-certik-tensorlean-formal-verification-gpu-kernels/)*

Researchers from Columbia University and CertiK built TensorLean, which uses the Lean proof assistant to verify production GPU kernels at the PTX (Nvidia's low-level instruction set) level. It checked 149 kernel instances, proved 83 correct, and found 38 bugs in libraries including FlashAttention and DeepGEMM.

**Why it matters:** Kernels like FlashAttention sit underneath most training and inference stacks, so silent bugs there can quietly corrupt results at scale. Verifying the compiled PTX rather than source code catches compiler-level issues too. Together with the Riemann proof check, it signals formal methods moving from academic niche into practical AI infrastructure.

**Topics:** research, chips

---
Read the original: https://www.marktechpost.com/2026/11/21/columbia-certik-tensorlean-formal-verification-gpu-kernels/
Canonical: https://www.parallelquant.com/posts/tensorlean-formally-verifies-gpu-kernels-finds-38-bugs-f452b8
Published by Parallel Quant — https://www.parallelquant.com
