samlobel

Thank you!

In this paper, the VAE also has a de-noising component. So, first you apply dropout to the input, and then try to recreate the un-corrupted input. That way the model...

I think it would be helpful to be able to use a local file instead of a remote_env file, in exactly the same way. That way you could edit it...

I haven't run it, but I imagine that it uses a different repository on this account, [dataset](https://github.com/yifanclifford/dataset). Unfortunately it doesn't provide the input files, but those look publicly available.

Thanks @yifanclifford , the original files look very helpful. Would you be able to point me to where you downloaded the raw data used in this paper? And, was I...

Hi, did you make any headway on this? Running into the same thing/ I saw you commented on a [related launchpad issue](https://github.com/deepmind/launchpad/issues/28) since your last update. Thanks!

Thanks for the response! Seems like in our case, at the very least there are some inconsistencies with how timelimits are being passed around by acme that we need to...

Hi, would you mind sharing more of what you did to get InferenceServer working? Think I'm treading the same path as you. I made the change to `timeout` which got...

Cool, thanks a bunch! I ended up with a similar solution, posting here in case its helpful to you or the next person. Uses [this gist](https://gist.github.com/willwhitney/dd89cac6a5b771ccff18b06b33372c75) that stacks arbitrary pytrees,...

We're getting slightly better performance with the inference_server compared to without (3800 FPS vs 2200 FPS, as well as slightly better training steps/second). The key was to instantiate multiple copies...