Efficient Long-Context Inference with SPEED Method
Discover SPEED's innovative layer-asymmetric KV visibility for faster, resource-efficient long-context inference in decoder-only language models.
New Kernel Framework for Safety Certification in Systems
Discover a novel kernel embedding framework that improves safety certification for dynamical systems by reducing errors and handling non-Markovian dynamics...
VibeServe: AI Agents Build Custom LLM Serving Systems
Discover how VibeServe uses AI agents to create bespoke LLM serving systems, optimizing performance for unique AI workloads and hardware setups.
Visual Fingerprints for Comparing LLM Outputs
Discover how visual fingerprints help compare large language model outputs, improving prompt design and model evaluation effectively.
Novelty-Based Tree-of-Thought Search for LLM Planning
Discover how Novelty-based Tree-of-Thought Search improves LLM reasoning and planning by enhancing efficiency and reducing resource use.