Search papers, labs, and topics across Lattice.
1
0
3
Stop letting your GPUs idle – a new tokenizer accelerates byte-level BPE on GPUs, achieving up to 7.6x speedup over HuggingFace's CPU implementation for long contexts.