Is there a mistake in computing the topk_maxmum ?
Opened this issue · 1 comments
wuzhenyubuaa commented
Hi, topk_maxmum, _ = mask_d.topk(inactive_channels, dim=1, largest=False, sorted=False)
, (in 120 lines, layers.py)
the author wants to obtain the top k values, however, the above code is not. if there are any errors?
hellozhuo commented
In the original code, we use topk to find the k smallest values in line 120, and mask out those values (assign them to 0) in lines 121-123.