-
Hi, thanks for contributing a great library!
I've been doing a close-up study of the `MultiheadAttentionPruner` implementation, and I have some concerns.
The pruning of output channel in out_pro…
-
Pruning a model with GLU results in an error when finding importance. GLU does not have any params but halves the input (in the given dimension). This is not accounted for during tracing, assigning in…
-
After compressing by KSE, the network is 2D kernels sparse, rather than channels sparse(obtained by the real channel pruning). Therefore, special computation hardware is needed to gain practical accel…
-
This is to report unexpected behavior where the broker selects too many segments to query. The identified scenario is when secondary partition information exists (eg, for range partitioning) but is n…
-
I'm trying to apply Channelwise Pruning. There persists few issues,
1. However, a few layers were not able to be pruned; the list of layers is printed out via the try and except call!
2. While loa…
-
### Environment
Any featurettes
### Description
Beq Janus notes: In the current implementation of the vocache, the generic extras file has no extras support in killObject and has no pruning/d…
-
Hi, thanks for the great work. I have a question about the experiments in predefined structured pruning methods. I am not sure I am understanding the paper correctly.
For predefined structured prun…
-
Not sure if this is intended behavior, but it looks like there might be an issue with concatenation based on the following test.
Code:
```
class TestModule(nn.Module):
def __init__(self, in_…
-
I used VGG19 trained for 160 epochs.
Then started pruning VGG19 with ratio
0.7,==> no of channels=0
0.2,==> no of channels=0
0.15=>no of channels>0, but accuracy is only 28%.
In readme.txt, th…
-
The following code snippet makes me puzzled. I know it is used to select `c_new` channels from c_in. WHY the lbound, rbound for channels and left, right for `alpha` are adjusted as you write. The orig…