BitBLAS is a library to support mixed-precision matrix multiplications, especially for quantized LLM deployment.
190
stars
21
forks
source link
[Dev] Refactor the range of INT Format to (-max_int_value - 1, max_int_value) #15
Closed
LeiWang1999 closed 2 months ago
Align the BitNET range:
refactor the int4 to (-8, 7) refactor the int2 to (-2, 1) refactor the int1 to (-1, 1)