ucbrise / actnn

ActNN: Reducing Training Memory Footprint via 2-Bit Activation Compressed Training

ucbrise/actnn Issues

the memory consume question
Closed 7 months ago3
how to avoid memory fragmentation in ActNN?
Closed 2 years ago2
Cannot save memory during the FP
Closed 2 years ago3
How to build as wheel ?
Closed 2 years ago
How i can remove this package?
Closed 2 years ago
Does this work well for Transformers?
Closed 2 years ago7
problem with torch._six on latest pytorch
Closed 2 years ago1
Does this work for any model with activation function as relu?
Updated 2 years ago4
AttributeError: 'QModule' object has no attribute 'stride'
Closed 2 years ago1
Environment configuration is too difficult
Closed 2 years ago
When I intalling actnn on ubuntu18.04, something wrong with it.
Updated 2 years ago3
I used Actnn in Deeplab v3 plus ResNet50, and the it degrade the iou by %2
Updated 3 years ago9
Could you add some comments about Qcheme.py?
Updated 3 years ago
How Actnn used in windows 10?
Updated 3 years ago
There is something wrong with loss.backward()
Updated 3 years ago2
A kind suggestion on the pytorch version.
Closed 3 years ago1
How to use it for GANs
Updated 3 years ago1
Installing actnn with some errors
Closed 3 years ago3
problem for unsuccessful compilation
Closed 3 years ago
[Feature] QDropout Implementation
Closed 3 years ago
Transformer Benchmarks?
Closed 3 years ago1
not support ceil_mode
Updated 3 years ago1
how to deploy actnn by libtorch?
Closed 3 years ago1
install problem
Closed 3 years ago1
RuntimeError: Expected kernel_size[0] * kernel_size[1] < 16 to be true, but got false. (Could this error message be improved? If so, please report an enhancement request to PyTorch.)
Closed 3 years ago2
There are some errors when I install actnn
Closed 3 years ago2
QConv1d: no valid convolution algorithms available in CuDNN
Closed 3 years ago3
what is pipeline_threshold used for?
Closed 3 years ago2