QuantTensor/test.ipynb

In [ ]:

import lib

import torch

T = torch.rand(8, 4)
Q = lib.quantize_q4_1(T)

dif = T - Q.dequantize()
print(abs(dif.sum()))

print(Q.nbytes)

print(Q.device)

Q.to("cuda")
print(Q.device)

tensor(0.0140)
20
cpu
cuda