Search papers, labs, and topics across Lattice.
1
0
3
Train models on half the memory: FlashOptim slashes optimizer memory requirements by over 50% without sacrificing accuracy, even when fine-tuning Llama-3.1-8B.
