Search papers, labs, and topics across Lattice.
4
0
11
20
ToolMaze reveals that LLMs suffer a staggering 37% drop in recovery performance when faced with implicit semantic failures, highlighting a critical vulnerability in current models.
A lightweight architecture that distills long textual sequences using visual tokens as dynamic queries boosts LLM performance on 2D table understanding by 23.9%.
Decoding LLMs with dynamic adapters doesn't have to be 2.5x slower: AdaFuse slashes latency by 2.4x with token-level pre-gating and fused kernel optimization.
GUI agents struggle with semantically ambiguous actions, but HATS tackles this by iteratively exploring hard cases and refining instruction alignment, leading to significant performance gains.
