-
Currently, the DTensor tensor subclass manages a `_local_tensor` attribute that represents the local tensor on the given rank. For efficient all-gather/reduce-scatter, we prefer to have a padded local…
awgu updated
9 months ago
-
I'm trying to use [xtensor][1] for defining an image class with extra header information. The example [Structures that embed shape and strides][2] from the documentation page "Extending xtensor" comes…
-
For multi-return functions (functions that return multiple tensors), one reasonable invariant one might expect for the keyword argument name for passing in an explicit out tensor to match the named fi…
-
### 🐛 Describe the bug
I'm trying to follow the instructions to efficiently load Hugging Face models from [`torchtitan`'s docs for FSDP1 -> FSDP2: Meta-Device Initialization](https://github.com/pyt…
-
In 'main.py' line 257 - 262, the author used the following codes to freeze the pruned weights:
for name, p in model.named_parameters():
if 'weight' in name:
tensor…
-
Currently they're defined like:
```
.op("quantized::add_out(Tensor qa, Tensor qb, Tensor(a!) out)"
"-> Tensor(a!) out",
c10::RegisterOperators::options()
.aliasAnalysis(at::Alias…
-
Thanks for the great work, but when I run the code, it shows the following error:
`File "/usr/local/lib/python3.9/dist-packages/torchvision/models/detection/anchor_utils.py", line 63, in AnchorGene…
-
The `__getitem__` of the datasets can not handle a namedtuple when there are multiple parallel workers. The parallel workers are needed to reach high data loading speeds for powerful compute nodes.
T…
-
I am trying to build models with multiple inputs and multiple outputs. A small example is created as below:
```
class Net(nn.Module):
def __init__(self):
nn.Module.__init__(self)
…
-
Hello
first of all thank you for your great work
i would like to extract the cross attention maps to visualize spatial attention in a synchronized way to my images (during training on my val and a…