OptiML: An End-to-End Framework for Program Synthesis and CUDA Kernel Optimization

Arijit Bhattacharjee, Heng Ping, Son Vu Le, Paul Bogdan, Nesreen K. Ahmed, Ali Jannesari · 2026

Read OptiML: An End-to-End Framework for Program Synthesis and CUDA Kernel Optimization by Arijit Bhattacharjee, Heng Ping, Son Vu Le, Paul Bogdan, Nesreen K. Ahmed, Ali Jannesari (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-12305.txt.

Description

Writing fast CUDA kernels means searching a huge space of low-level tweaks under noisy, costly hardware feedback. Arijit Bhattacharjee led this work with five co-authors, presenting OptiML, an end-to-end frame that turns plain language or input CUDA code into performance-tuned kernels. Optimisation is treated as search under verification, not as one-shot generation.

When input is language, a Mixture-of-Thoughts generator proposes a first runnable kernel. A search optimiser then refines kernels by Monte Carlo tree search over model-driven edits, guided by profiler rewards, with each candidate compiled, checked, and profiled. Tests show verified gains over strong language-model baselines, with clear optimisation paths grounded in profiler evidence rather than guesswork.

Catalog record

CREATORArijit Bhattacharjee, Heng Ping, Son Vu Le, Paul Bogdan, Nesreen K. Ahmed, Ali Jannesari
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE95 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-12305

Browse the full library · About the Tower of Babel