Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
4
0
4
0
A single attention module dynamically switches between linear prefilling and full decoding, achieving $21.9\times$ faster prefilling and Transformer-level long-context recall without the quadratic bottleneck.
Closing nearly 90% of performance gaps in low-bit quantized models with a compact repair codec could redefine efficiency in language model deployment.
Multi-turn jailbreaks exploit user intent in ways that traditional safety measures fail to detect, revealing a critical vulnerability in LLM interactions.
LLM defenses can achieve a 79% reduction in attack success rate against evolving multi-round attacks by using a stateful, multi-agent cooperative framework.