Search papers, labs, and topics across Lattice.
2
0
4
3
Context-aware embeddings can drastically improve multimodal document retrieval, revealing insights that independent page evaluations miss.
Overcome the quadratic attention bottleneck in vision-language models with Parallel-ICL, a method that achieves comparable performance to full-context learning while drastically reducing inference time.
