HaiAI123

Curated Global AI Tools Directory

NewIndustry update
98
heat index
New

Columbia and CertiK launch TensorLean for formal GPU kernel verification

11/22/2026 — 11/22, 10:18·1 sources (all)·1 reports (all)

Key takeaways

  • Researchers from Columbia University and CertiK proposed TensorLean, a formal verification method for production GPU kernels at the PTX leve…

Background & analysis

On November 22, 2026, MarkTechPost reported that a research team from Columbia University and CertiK has introduced TensorLean, a tool that applies formal verification to production GPU kernels. Working at the PTX level, TensorLean uses Lean to check whether the actual behavior of these kernels matches the intended specifications, rather than relying on conventional testing to spot defects after the fact.

According to the report, the method was applied to a set of real-world kernels and evaluated on 149 kernel instances in total. Of those instances, 83 were formally proven correct. The same verification effort also surfaced 38 bugs across kernels including FlashAttention and DeepGEMM, which are used in performance-critical computation in real deployments.

The available material consists of this single report, so further details on how TensorLean is implemented, the scope of kernels it can cover, and any subsequent development or adoption plans are not documented in the sources provided. The findings described remain as presented by MarkTechPost, with no additional independent confirmation available at this point.

AI-generated from 1 reports · updated 28 minutes ago

updream
24-hour heatHeat index 98
Just charted — trend is building up
24 hours agonow

Reports on this story headlines open the original

Today
  1. Researchers from Columbia University and CertiK introduced TensorLean, which formally verifies production GPU kernels at the PTX level using Lean. It checked 149 kernel instances, proved 83 correct and, according to MarkTechPost, surfaced 38 bugs in kernels including FlashAttention and DeepGEMM.

    MarkTechPostAI score 61

Other stories people are talking about

How is heat calculated?About the method

Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.

This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.

Surge
Discussion rising fast
New
First report within 6 hours
Rising
Still gathering discussion

Back to AI industry updates →