float16

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP

🎯 Gradient Accumulation for TensorFlow 2

Convert a double-precision floating-point number to the nearest half-precision floating-point number.

Convert a half-precision floating-point number to the nearest double-precision floating-point number.

Return an integer corresponding to the unbiased exponent of a half-precision floating-point number.

Return an integer corresponding to the significand of a half-precision floating-point number.

Return a boolean indicating if the sign bit for a half-precision floating-point number is on (true) or off (false).

A deep learning framework built to be read: lazy tensors, reverse-mode autograd, kernel fusion, a printable IR, conv layers, dtypes from int8 to float64, and JIT backends for C and CUDA.

Create a half-precision floating-point number from an unsigned integer corresponding to an IEEE 754 binary representation.