Torsten Scholak

no, no, you are right, it would be for some unknown, but specific `a` and `b`. implementing the following is not possible with what `Hedgehog.Function` provides today: ``` mkGen ::...

I was testing with shapeless 2.3.1, typelevel scalac 2.11.8, and fs2-cassandra 0.1.6. You can easily reproduce the problem by adding `scalaOrganization := "org.typelevel"` and `"-Yliteral-types"` to fs2-cassandra's `build.sbt`.

and yes, `-Yliteral-type` activates Miles' implementation of SIP-42, https://github.com/scala/scala/pull/5310

Hi @goldfirere, thank you for this very detailed response and the great ideas! I think I like version 2 (without `Proxy`) better. A small issue is that the Haskell Language...

Hi @GuoZhilong, thanks for checking out our code. We are using pipenv to manage the development environment. Therefore, you will find the package requirements in the file https://github.com/ElementAI/duorat/blob/master/Pipfile. I hope...

Hi @sanjayss34, thanks for checking out our code! I cannot make a promise right now, but we're looking into it!

Hi @sanjayss34, we are using Pytorch’s automatic mixed precision mode, which was introduced in 1.6 and originates from NVidia’s own APEX project. I recommend you read up on it here:...

with regards to where and how the automatic mixed precision is implemented in duorat, please have a look at the training loop, specifically, around here: https://github.com/ElementAI/duorat/blob/master/scripts/train.py#L275

Hi @hclent, thanks for checking out our code! The problem you describe seems to be related to the fact that the `data` directory is empty. `../../data/train.libsonnet` is a jsonnet file...

Hi, that means that the picard backend is either not running (yet) or unresponsive. Does this happen reproducibly?