Search papers, labs, and topics across Lattice.
Massachusetts Institute of Technology
1
0
SoftWater slashes quantization error by up to 8.3x while maintaining near-lossless performance in language models, revolutionizing softmax layer efficiency.
