Search papers, labs, and topics across Lattice.
5
0
8
17
SKILLER achieves up to 20.4 percentage points improvement in skill generation for small language models, making high-quality task execution accessible without the prohibitive costs of closed-source solutions.
Achieving an 81.3% Print-Ready Rate, PosterMELD outperforms existing poster generation methods while retaining full editability and design diversity.
OmniCoT recalibrates the challenges of panoramic reasoning, enabling MLLMs to leverage global evidence for complex multi-step inference.
ReMMD-Agent achieves a remarkable 41.80% accuracy in detecting misinformation across complex multilingual and multi-image scenarios while slashing verification costs by up to 80%.
Forget bigger models: massive gains in document parsing accuracy are still possible through smarter data engineering alone.
