Search papers, labs, and topics across Lattice.
1
0
3
Forget fine-tuning: surprisingly, single neuron activations in VLMs can be directly probed to create classifiers that outperform the full model, with 5x speedups.
