Search papers, labs, and topics across Lattice.
3
0
4
Latent Memory slashes token usage by up to 10x while maintaining competitive performance in multimodal question answering.
Smooth gradient adjustments in DRPO prevent harmful policy shifts, leading to more stable and efficient LLM training.
VLMs can't play the shell game: they fail to track visually identical objects over time, but a simple fine-tuning strategy using generated trajectories can fix this.
