Expand description
Sub-byte 4-bit (s4) quantization operations and layers.
Functionsยง
- convolve_
s4 - Performs per-tensor quantized int4 (
s4) 2D Convolution layer. - fully_
connected_ s4 - Performs per-tensor quantized int4 (
s4) Fully Connected layer. - pack_
s4_ pair - Packs two 4-bit signed integers
(low, high)in the range[-8, 7]into a single byte. - unpack_
s4_ pair - Unpacks a byte containing two 4-bit signed integers into
(low_nibble, high_nibble).