shayne-longpre

@takiholadi I'd recommend updating to the latest nightly versions of seqio and tfds as bugs in their older versions affected our repo. I will look into what the best stable...

@danczs Thanks for the question. A couple thoughts: * I'm surprised that the SirNeural did so well. The reason we recommended Enrico's version is because it properly applies the dataset...

@danczs Hmm I'm not sure why it was so low. I noticed that a few recent papers seem to have gotten strong results with a 100k sample of the training...

@takiholadi Yes, this looks correct!

@quq99 it is quite memory intensive. We ran it a while ago internally with Google infrastructure so I don't have specific numbers unfortunately, but in terms of compute it should...

You can also now manually download the Dialog submixture (and the others) -- see the new README! :)

Here are some resources, and there should be more info in the documentation: https://github.com/google/seqio#optional-offline-caching. This caching is for if you are using the same vocabulary as T5. If you want...

@quq99 Good question. As the Flan Collection (or P3, or Natural Instructions v2) is a compilation of hundreds of different datasets, with many different licenses, the rendered data would not...

@quq99 Update: we plan to release this in the last week of May.

@balachandarsv apologies again for the wait on this. It turns out license labelling is much more complex than we had originally anticipated. It has gone from a side project into...