Adaptive Self-improvement LLM Agentic System for ML Library Development

Genghan Zhang, Weixin Liang, Olivia Hsu, Kunle Olukotun
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:75427-75452, 2025.

Abstract

ML libraries, often written in architecture-specific programming languages (ASPLs) that target domain-specific architectures, are key to efficient ML systems. However, writing these high-performance ML libraries is challenging because it requires expert knowledge of both ML algorithms and the ASPL. Large language models (LLMs), on the other hand, have shown general coding capabilities. However, challenges remain when using LLMs for generating ML libraries using ASPLs because 1) this task is complicated even for human experts and 2) there are limited code examples due to the esoteric and evolving nature of ASPLs. We present an adaptive self-improvement agentic system that enables LLMs to perform such complex reasoning under limited data by iteratively improving their capability through self-generated experience. In order to evaluate the effectiveness of our system, we construct a benchmark of a typical ML library and generate ASPL code with both open and closed-source LLMs on this benchmark. Our results show improvements of up to $3.9\times$ over a baseline single LLM.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-zhang25at, title = {Adaptive Self-improvement {LLM} Agentic System for {ML} Library Development}, author = {Zhang, Genghan and Liang, Weixin and Hsu, Olivia and Olukotun, Kunle}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {75427--75452}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/zhang25at/zhang25at.pdf}, url = {https://proceedings.mlr.press/v267/zhang25at.html}, abstract = {ML libraries, often written in architecture-specific programming languages (ASPLs) that target domain-specific architectures, are key to efficient ML systems. However, writing these high-performance ML libraries is challenging because it requires expert knowledge of both ML algorithms and the ASPL. Large language models (LLMs), on the other hand, have shown general coding capabilities. However, challenges remain when using LLMs for generating ML libraries using ASPLs because 1) this task is complicated even for human experts and 2) there are limited code examples due to the esoteric and evolving nature of ASPLs. We present an adaptive self-improvement agentic system that enables LLMs to perform such complex reasoning under limited data by iteratively improving their capability through self-generated experience. In order to evaluate the effectiveness of our system, we construct a benchmark of a typical ML library and generate ASPL code with both open and closed-source LLMs on this benchmark. Our results show improvements of up to $3.9\times$ over a baseline single LLM.} }
Endnote
%0 Conference Paper %T Adaptive Self-improvement LLM Agentic System for ML Library Development %A Genghan Zhang %A Weixin Liang %A Olivia Hsu %A Kunle Olukotun %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-zhang25at %I PMLR %P 75427--75452 %U https://proceedings.mlr.press/v267/zhang25at.html %V 267 %X ML libraries, often written in architecture-specific programming languages (ASPLs) that target domain-specific architectures, are key to efficient ML systems. However, writing these high-performance ML libraries is challenging because it requires expert knowledge of both ML algorithms and the ASPL. Large language models (LLMs), on the other hand, have shown general coding capabilities. However, challenges remain when using LLMs for generating ML libraries using ASPLs because 1) this task is complicated even for human experts and 2) there are limited code examples due to the esoteric and evolving nature of ASPLs. We present an adaptive self-improvement agentic system that enables LLMs to perform such complex reasoning under limited data by iteratively improving their capability through self-generated experience. In order to evaluate the effectiveness of our system, we construct a benchmark of a typical ML library and generate ASPL code with both open and closed-source LLMs on this benchmark. Our results show improvements of up to $3.9\times$ over a baseline single LLM.
APA
Zhang, G., Liang, W., Hsu, O. & Olukotun, K.. (2025). Adaptive Self-improvement LLM Agentic System for ML Library Development. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:75427-75452 Available from https://proceedings.mlr.press/v267/zhang25at.html.

Related Material