Quantize weights with 16 bits
4 files changed