Evolving Infra: Agent-Adaptive LLM Serving at Scale
cs.DC
Submitted: 2026-11-09
Updated: 2026-11-12
Code: https://github.com/mit-han-lab/EvolvingInfra
Terminology
Sources
- State of AI: An Empirical 100 Trillion Token Study with OpenRouter
- Longformer: The Long-Document Transformer
- DeepSeek-V3 Technical Report
- DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
- DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models
- DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
- Frontier: Towards Comprehensive and Accurate LLM Inference Simulation
- GLM-5: from Vibe Coding to Agentic Engineering
- semi-PD: Towards Efficient LLM Serving via Phase-Wise Disaggregated Computation and Unified Storage
- MemServe: Context Caching for Disaggregated LLM Serving with Elastic Memory Pool
- Mistral 7B
- Kimi K2.5: Visual Agentic Intelligence
- Continuum: Efficient and Robust Multi-Turn LLM Agent Scheduling with KV Cache Time-to-Live
- Agentic Coding in the Wild: Characterizing GitHub Copilot Traces at Production Scale
- xLLM Technical Report
- AlphaEvolve: A coding agent for scientific and algorithmic discovery
- Fast Transformer Decoding: One Write-Head is All You Need
- BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents
- Arrow: Adaptive Scheduling Mechanisms for Disaggregated LLM Inference Architecture
Related papers
- Programmable Probabilistic Computer with 1,000,000 p-bits
- BatchGen: An Architecture for Scalable and Efficient Batch Inference
- Seamless Reconfiguration for DAG-BFT (Technical Report)
- ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving
- iScheduler: Reinforcement Learning-Driven Continual Optimization for Large-Scale Resource Investment Problems
- TIRx: A Unified Tile Primitive ML Compiler Across Heterogeneous Backends