Many models have identical implementations of prune_heads it would be nice to store that implementation as a method on PretrainedModel and reduce the redundancy.

Hi, is there any plan to provide a tutorial of showing an example of employing the Transformer as an alternative of RNN for seq2seq task such as machine translation?

Is there an explanation for what these parameters represent?

octave_base_scale
scales_per_octave
anchor_ratios
anchor_strides
featmap_strides

And how can I calculate the best ones for my data? (Which contains lots of very small objects)

Environment:

1. Framework: PyTorch
2. Framework version: 1.3.1
3. Horovod version: 0.19.0
4. MPI version: 4.0.2
5. CUDA version: N/A
6. NCCL version: N/A
7. Python version: 3.7.5
8. OS and version: Mac OS 10.15.2
9. GCC version: 9.2.0

Checklist:

1. Did you search issues to find if somebody asked this question before? Yes
2. If your question is about hang, did you read [this d

I tried selecting hyper parameters of my model following "Tutorial 8: Model Tuning" below:
https://github.com/flairNLP/flair/blob/master/resources/docs/TUTORIAL_8_MODEL_OPTIMIZATION.md

Although I got the "param_selection.txt" file in the result directory, I am not sure how to interpret the file, i.e. which parameter combination to use. At the bottom of the "param_selection.txt" file, I found "

Describe the bug

Calling Predictor.get_gradients() returns an empty dictionary

To Reproduce
I am replicating the binary sentiment classification tasked described in the paper 'Attention is not Explanation ' (Jain and Wallace 2019 - https://arxiv.org/pdf/1902.10186.pdf).

My first experiment is on the Stanford Sentiment TreeBank Dataset. I need to measure the correlation between th

Several parts of the op sec like the main op description, attributes, input and output descriptions become part of the binary that consumes ONNX e.g. onnxruntime causing an increase in its size due to strings that take no part in the execution of the model or its verification.

Setting __ONNX_NO_DOC_STRINGS doesn't really help here since (1) it's not used in the SetDoc(string) overload (s

Recently, when I training NAT models with fairseq, I encountered many problems, such as segment fault, CUDA_error etc., I checked a lot of issues, but found that there is no right solution under the master branch.

After kinds of exploration, I finally installed it correctly. You can reproduce it

The documentation about edge orientation is inconsistent. In the Creating Message Passing Networks tutorial, the main expression says that e𝑖,𝑗 denotes (optional) edge features from node 𝑖 to node 𝑗., the attached expression also suggests it. However, in documentation to MessagePassing.message(), the documentation says Constructs messages from node 𝑗 to node 𝑖 (this is actually true).

I

Describe the bug
I try to run tensorboardX/examples/demo_graph.py for jupyter notebook (launched by anaconda navigator) and I get the error seen at Additional context.

I just copy paste the code to notebook from Github.

Minimal runnable code to reproduce the behavior
class SimpleModel(nn.Module):
def init(self):
super(SimpleModel, self).init()

Similar to the tutorial on custom losses in SVI, we should have a tutorial on implementing custom MCMC kernels using the new MCMC API. Something simple like SGLD seems like a good starting point.

There is a problem with padding on line 73-75 . What if the sentence length is larger than maxlen? Then we end up with sequences of varying length and [line 214](http

Raised by #2337

Is your feature request related to a problem? Please describe.

OpenMined/PySyft#2498

Describe the solution you'd like

Solution: extend nn.Module() with a .parameters_and_statistics() iterator which will look for this kind of information so that we can make sure that federated learning properly moves everything from machien to machine.

**Describe altern

Can someone explain how dimensions of the anchor boxes are calculated from anchor ANCHOR_SCALES and ANCHOR_RATIOS? How do they relate to generating 1:1, 1:2 or 2:1 aspect ratio anchor boxes with box areas 128^2, 256^2 as mentioned in the Faster RCNN paper?

Sorry to bother you.

🚀 Feature

we shall also cover TPU usage as we are supporting it

Motivation

now all changes are tested for GPUs and CPU but we do not have a check for TPU yet

Pitch

getting coverage back to ~99%