Search papers, labs, and topics across Lattice.
1
0
3
Speech models can now be quantized to INT4 with near-lossless performance thanks to a new evolution strategy-based calibration method tailored for audio activations.
