Search papers, labs, and topics across Lattice.
1
0
3
6
Ditch the black box: this new lossless compression method uses RL to train a Transformer to compress data into *discrete* token sequences, boosting compression ratios while preserving semantic structure.
