PassNet: Scaling Large Language Models for Graph Compiler Pass Generation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Yiqun, Wu, Yingsheng, Yang, Ruqi, Zheng, Enrong, Qiu, Honglei, He, Sijun, Liang, Tai, Wu, Jingjing, Zhou, Yuhan, Zhang, Yiwei, Chen, Dongyan, Yi, Weihan, Li, Xinqi, Bao, Siqi
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914612241235968
author Liu, Yiqun
Wu, Yingsheng
Yang, Ruqi
Zheng, Enrong
Qiu, Honglei
He, Sijun
Liang, Tai
Wu, Jingjing
Zhou, Yuhan
Zhang, Yiwei
Chen, Dongyan
Yi, Weihan
Li, Xinqi
Bao, Siqi
author_facet Liu, Yiqun
Wu, Yingsheng
Yang, Ruqi
Zheng, Enrong
Qiu, Honglei
He, Sijun
Liang, Tai
Wu, Jingjing
Zhou, Yuhan
Zhang, Yiwei
Chen, Dongyan
Yi, Weihan
Li, Xinqi
Bao, Siqi
contents Modern tensor compilers such as TorchInductor deliver substantial speedups on mainstream models, yet face a systematic performance ceiling on long-tail workloads -- our profiling shows that 43% of real-world subgraphs experience end-to-end slowdowns under default compilation. While LLMs offer a path toward automated optimization, existing efforts focus on standalone kernel generation. We argue that pass generation -- where LLMs author structured graph transformations that integrate directly into compiler pipelines -- is the more appropriate abstraction. We propose PassNet, the first large-scale ecosystem for LLM-based compiler pass generation, comprising: (1) PassNet-Dataset, over 18K unique computational graphs from 100K real-world models; and (2) PassBench, 200 curated long-tail fusible tasks (comprising 2,060 subgraphs in total) evaluated under the Error-aware Speedup Score (ES_t) -- a metric unifying correctness, stability, and performance -- with layered integrity defenses against systematic LLM exploitation. Experiments reveal that PassBench is both highly discriminative and genuinely unsaturated: the best frontier model trails TorchInductor by 37% in aggregate, yet on individual subgraphs LLMs achieve up to 3x speedup over the same compiler -- indicating that the bottleneck is consistency, not capability. Fine-tuning a small model on merely ~4K PassNet trajectories yields a 2.67x improvement approaching frontier-model performance, demonstrating substantial headroom and validating PassNet as live training infrastructure for advancing LLM-driven compiler optimization. All data, benchmarks, and tooling are publicly available.
format Preprint
id arxiv_https___arxiv_org_abs_2605_29357
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle PassNet: Scaling Large Language Models for Graph Compiler Pass Generation
Liu, Yiqun
Wu, Yingsheng
Yang, Ruqi
Zheng, Enrong
Qiu, Honglei
He, Sijun
Liang, Tai
Wu, Jingjing
Zhou, Yuhan
Zhang, Yiwei
Chen, Dongyan
Yi, Weihan
Li, Xinqi
Bao, Siqi
Artificial Intelligence
Machine Learning
Programming Languages
Modern tensor compilers such as TorchInductor deliver substantial speedups on mainstream models, yet face a systematic performance ceiling on long-tail workloads -- our profiling shows that 43% of real-world subgraphs experience end-to-end slowdowns under default compilation. While LLMs offer a path toward automated optimization, existing efforts focus on standalone kernel generation. We argue that pass generation -- where LLMs author structured graph transformations that integrate directly into compiler pipelines -- is the more appropriate abstraction. We propose PassNet, the first large-scale ecosystem for LLM-based compiler pass generation, comprising: (1) PassNet-Dataset, over 18K unique computational graphs from 100K real-world models; and (2) PassBench, 200 curated long-tail fusible tasks (comprising 2,060 subgraphs in total) evaluated under the Error-aware Speedup Score (ES_t) -- a metric unifying correctness, stability, and performance -- with layered integrity defenses against systematic LLM exploitation. Experiments reveal that PassBench is both highly discriminative and genuinely unsaturated: the best frontier model trails TorchInductor by 37% in aggregate, yet on individual subgraphs LLMs achieve up to 3x speedup over the same compiler -- indicating that the bottleneck is consistency, not capability. Fine-tuning a small model on merely ~4K PassNet trajectories yields a 2.67x improvement approaching frontier-model performance, demonstrating substantial headroom and validating PassNet as live training infrastructure for advancing LLM-driven compiler optimization. All data, benchmarks, and tooling are publicly available.
title PassNet: Scaling Large Language Models for Graph Compiler Pass Generation
topic Artificial Intelligence
Machine Learning
Programming Languages
url https://arxiv.org/abs/2605.29357