SafeHarbor: Advanced Memory Guardrail for LLM Safety
Discover SafeHarbor, a dynamic memory-augmented framework enhancing LLM agent safety while maximizing utility and minimizing harmful task risks.
Optimizing LLM Multi-Agent Communication with Active Learning
Discover how active learning enhances communication structure in LLM multi-agent systems, boosting performance and reducing token use efficiently.
Empirical Study on Proactive Coding Assistants in Software
Explore real-world developer behavior and the limits of simulations in proactive coding assistants with insights from a large-scale empirical study.
int4 KV Cache Beats fp16 on Apple Silicon: Faster AI Performance
Discover how int4 KV cache outperforms fp16 on Apple Silicon, boosting AI model speed and efficiency with minimal quality loss and advanced quantization.
Efficient Transformers with Budgeted Attention Allocation
Optimize transformer models using Budgeted Attention Allocation for flexible, cost-effective compute control and improved efficiency in NLP tasks.