Haisheng Chen
Haisheng Chen
I met a problem when I try to compile the WQLinear forward function. ``` @torch.no_grad() def test_WQLinear(): from rotquant.modules.qmodule import WQLinear linear = nn.Linear(4096, 4096, bias=False).cuda().half() w = torch.randn_like(linear.weight.data) w,...
In this approach, I follow a method similar to PyTorch's QConv1dInt8 implementation. First, I use an unsqueeze operation on the input tensor along the second dimension. Next, I borrow some...
I will look into this
@mvafin I adapted the translate_quantized_convnd_base to conv1d and merged the test function to test_quantized_convnd.py.
Since the support for quantized_conv1d_relu is in another issue, I comment it out.