@mylifcc: Major Experiment: Using LLM as an 'Optimization Agent' for Automatic Loop Scheduling! Just read this paper accepted at PACT 2025: 'Agentic Auto-Scheduling: An Experimental Study of LLM-Guided Loop Opt…'
Summary
Introduces a paper accepted at PACT 2025, proposing the ComPilot framework, which uses off-the-shelf LLMs as optimization agents to automatically optimize complex loop nests without fine-tuning, achieving a geometric mean speedup of 3.54x, surpassing the SOTA Pluto.
View Cached Full Text
Cached at: 06/28/26, 06:12 PM
Groundbreaking Experiment: Let LLM Be an “Optimization Agent” to Automatically Schedule Loops!
Just finished reading this paper, recently accepted at PACT 2025: Agentic Auto-Scheduling: An Experimental Study of LLM-Guided Loop Optimization
Core contribution: Proposing the ComPilot framework — a zero-shot, fine-tuning-free off-the-shelf LLM that, through closed-loop interaction with the compiler, automatically optimizes complex loop nests, achieving up to 3.54× geometric mean speedup, and even surpassing the SOTA Pluto in most cases!
This is not simply “letting LLM write code,” but a true new paradigm of Agentic Auto-Scheduling.
Similar Articles
@vintcessun: LLMs can now write their own work scripts, decomposing tasks into a group of sub-agents that execute in parallel. A single assistant thinks sequentially, and when faced with tasks like codebase audits or large-scale refactoring, it's either slow or prone to mixing up its thoughts. pi-dynamic-workflows lets the model directly generate a JS script, using agent() and parallel() for task orchestration, runs them in a sandbox, and then aggregates the results—with real-time progress display. The essence is transforming "one person working" into "one person writing a scheduling script, while minions execute in parallel."
Introducing pi-dynamic-workflows, a tool that enables LLMs to dynamically orchestrate multiple sub-agents for parallel task execution by generating JavaScript scripts, suitable for code audits, large-scale refactoring, and similar scenarios.
LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling
This paper introduces AutoTTS, an environment-driven framework that automates the discovery of test-time scaling strategies for LLMs by formulating it as controller synthesis. It demonstrates improved accuracy-cost tradeoffs on mathematical reasoning benchmarks with minimal computational overhead.
AccelOpt: A Self-Improving LLM Agentic System for AI Accelerator Kernel Optimization
AccelOpt is a self-improving LLM agentic system that autonomously optimizes AI accelerator kernels through iterative generation and optimization memory, achieving 49-61% peak throughput improvements on AWS Trainium while being 26x cheaper than Claude Sonnet 4.
AutoLLMResearch: Training Research Agents for Automating LLM Experiment Configuration -- Learning from Cheap, Optimizing Expensive
This paper introduces AutoLLMResearch, an agentic framework that automates the configuration of expensive LLM experiments by learning from low-fidelity environments and extrapolating to high-cost settings. It aims to reduce computational waste and reliance on expert intuition in scalable LLM research.
@Xudong07452910: Many people's default habit when using AI coding is: go straight to the strongest model. For the same task, should Sonnet or Opus do it? Most of the time this decision is made on a whim. So this paper Agent-as-a-Router raises a very practical question: if different models excel at different tasks…
This paper proposes the Agent-as-a-Router framework, which transforms model routing into a dynamic, iterative process. Based on task type and real-time execution feedback, it selects the most suitable LLM to improve coding performance and cost efficiency.