Optimizing PyTorch Inference with LLM-Based Multi-Agent Systems

2025/11/21 by Kirill Nagaitsev, Nagaitsev, Kirill, Luka Grbčić +5
Computer Science · #Parallel Computing and Optimization Techniques #Advanced Neural Network Applications #Embedded Systems Design Techniques

paper · pdf · doi:10.48550/arxiv.2511.16964

Abstract

Artifact for: Optimizing PyTorch Inference with LLM-Based Multi-Agent Systems

Citations

• Towards Robust Agentic CUDA Kernel Benchmarking, Verification, and Optimization
• Astra: A Multi-Agent System for GPU Kernel Performance Optimization
• SE-Agent: Self-Evolution Trajectory Optimization in Multi-Step Reasoning with LLM-Based Agents
• GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning
• CUDA-L1: Improving CUDA Optimization via Contrastive Reinforcement Learning
• Kevin: Multi-Turn RL for Generating CUDA Kernels
• GPU Kernel Scientist: An LLM-Driven Framework for Iterative Kernel Optimization
• AlphaEvolve: A coding agent for scientific and algorithmic discovery
• CUDA-LLM: LLMs Can Write Efficient CUDA Kernels
• GSO: Challenging Software Optimization Tasks for Evaluating SWE-Agents
• AgentNet: Decentralized Evolutionary Coordination for LLM-based Multi-Agent Systems
• KernelBench: Can LLMs Write Efficient GPU Kernels?
• ThunderKittens: Simple, Fast, and Adorable AI Kernels
• BudgetMLAgent: A Cost-Effective LLM Multi-Agent system for Automating Machine Learning Tasks
• EvoAgent: Towards Automatic Multi-Agent Generation via Evolutionary Algorithms
• Persistent memory for AI coding agents: a pre-registered SWE-bench Verified benchmark
• MLAgentBench: Evaluating Language Agents on Machine Learning Experimentation
• Large Language Model Cascades with Mixture of Thoughts Representations for Cost-efficient Reasoning
• AIKernel Semantic DSL Compiler and Deterministic Agent Execution Architecture
• MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework
• FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance
• FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
• cuDNN: Efficient Primitives for Deep Learning

Related

• InferenceBench: A Benchmark for Open-Ended LLM Inference Optimization by AI Agents
• Optima: Optimizing Effectiveness and Efficiency for LLM-Based Multi-Agent System
• MAPRO: Recasting Multi-Agent Prompt Optimization as Maximum a Posteriori Inference
• MARFT: Multi-Agent Reinforcement Fine-Tuning
• HiveMind: Contribution-Guided Online Prompt Optimization of LLM Multi-Agent Systems