Search papers, labs, and topics across Lattice.
3
0
6
Compositionality in neural networks only emerges in a narrow depth-connectivity regime, with specific architectural constraints dictating success or failure.
Standard Gaussian initialization can lead to mode collapse, but a new approach using guidance potential posteriors dramatically enhances image diversity.
Squeeze your LLM's KV cache by 82% without significant performance loss using VQKV's novel vector quantization approach.
