Module float_ops 

Floating-point (f32 and f16) fallback operations and layers.

Functions

convolve_f32

Performs 2D Convolution for 32-bit floating point tensors.

f16_to_f32

IEEE-754 16-bit half-precision floating-point conversion to 32-bit single precision float.

f32_to_f16

32-bit single precision float conversion to IEEE-754 16-bit half-precision float.

fully_connected_f32

Performs Fully Connected layer for 32-bit floating point tensors.

relu6_f32

In-place ReLU6 for 32-bit float buffer (min(max(val, 0.0), 6.0)).

relu_f32

In-place ReLU for 32-bit float buffer (max(val, 0.0)).

softmax_f32

Softmax for 32-bit floating point tensors.