Xuming Hu

4 papers The Hong Kong University of Science and Technology (Guangzhou)

• ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference
• Don't Just Chase “Highlighted Tokens” in MLLMs: Revisiting Visual Holistic Context Retention
• LoTA-QAF: Lossless Ternary Adaptation for Quantization-Aware Fine-Tuning
• RTV-Bench: Benchmarking MLLM Continuous Perception, Understanding and Reasoning through Real-Time Video