Mingxuan Yuan

7 papers Huawei Noah's Ark Lab

• Accurate KV Cache Eviction via Anchor Direction Projection for Efficient LLM Inference
• AttentionPredictor: Temporal Patterns Matter for KV Cache Compression
• Benchmarking End-To-End Performance of AI-Based Chip Placement Algorithms
• CORE: Collaborative Optimization with Reinforcement Learning and Evolutionary Algorithm for Floorplanning
• Dependency Matters: Enhancing LLM Reasoning with Explicit Knowledge Grounding
• High-Performance Arithmetic Circuit Optimization via Differentiable Architecture Search
• Preserving LLM Capabilities through Calibration Data Curation: From Analysis to Optimization