Accurate Neural Training with 4-bit Matrix Multiplications at Standard Formats

Open in new window