Eric Jang

Some discrepancies from the paper

2

For the Mini-ImageNet ResNet-based encoder, I believe the authors use mean pooling instead of max pooling, as done by this implementation. I am currently unable to reproduce the mini-imagenet results...

How does Captura/FFMpeg handle long recording sessions

1

FFMpegVideoWriter default settings does not seem to guarantee that encoding is "-tune ultrafast", so it's quite possible that the user is sending frames to the named output pipe faster than...

Support for Multiple Robots (Mechanisms?)

2

Hello, Great work with this suite of packages! I came across this yesterday and have been playing around with the notebooks to learn more about rigid body dynamics. It appears...

Add writer.from_iterable_dataset for writing IterableDatasets to .beton

1

The current normalizer transform when operating on GPU tensors uses cupy, which assumes device=0 unless otherwise stated. This results in a bug where trying to run ffcv pipelines with Normalize...

I have two .beton datasets with the same schema. how can I concatenate files or load two different beton files into one Loader instance?

support for static content / html middleware?

6

Hello, Do you plan on having support for serving html pages + other static content? Or is that something to be implemented in a separate Middleware Module?

"Super expression must either be null or a function" when trying to extend Renderer in test

1

Hi, thanks for putting this awesome repo together. I'm new to the Node/Babel/ES6/Mocha/Webpack ecosystem, so please bear with my novice question: I added a file `wiggle/test/client.js` with the following contents:...

reparameterization trick?

1

This is really cool! It looks like you were able to get good results by sampling `gen_z: np.random.uniform(-1., 1., size=(GENERATOR_BATCH,GENERATOR_SEED)).astype(np.float32)` on each train step without using the reparameterization trick, which...

gfx900 on ROCm 3.3 unexpectedly slower than P100 + CUDNN on DAWN benchmark

13

## 🐛 Bug The baseline code in [this colab](https://colab.research.google.com/github/davidcpage/cifar10-fast/blob/master/bag_of_tricks.ipynb#scrollTo=7WeTS0FF4iQE) provides a Pytorch example running on a V100 GPU with CUDNN to train a CIFAR10 classifier in 75 seconds. ``` epoch...

torch.cuda.memory_stats returns all 0s

2

## 🐛 Bug Calling `torch.cuda.memory_stats` on gfx900 GPU (Frontier Vega) or any of the methods in https://pytorch.org/docs/stable/cuda.html#memory-management results in 0s. ## To Reproduce Steps to reproduce the behavior: 1. Run...