Some tweaks to the Editor

[srush]

• Default sort by popularity and include number of prompts

• Global progress table

• Separated out new prompts and select, and the columns (I think this is okay, but might break)

• Added text wrapping so that you can view long prompts

• Added a template viewer section

[stephenbach]

Awesome! I love it!

I'm having some rendering issues for TREC in the template viewer:

I can look more into, but just flagging in case you know a quick fix.

More broadly, @VictorSanh added some template viewing capabilities to promptapp.py. Do we want to merge them together in this PR?

[VictorSanh]

Also, for the template viewer, I think it's more important to be able to scroll through a lot of examples for a given prompt than scrolling through a lot of prompts for a given example to quickly identify potential bugs (see example #36), thus my comment on overlapping global viewers

[srush]

@VictorSanh I think the reason I liked the viewer in this tool as well, was just to make sure that people don't overwrite the same prompt many times. Makes it easy to see what is there when adding a new prompt.

[srush]

@stephenbach here is what my side looks like.

Wonder if it is a wide / narrow screen thing.

(btw, this is an example, where I think having a filter might be better than a switch statement)

[VictorSanh]

@VictorSanh I think the reason I liked the viewer in this tool as well, was just to make sure that people don't overwrite the same prompt many times. Makes it easy to see what is there when adding a new prompt.

Right, during the sprint we won't have such a problem because one dataset = one person, so you won't have multiple people working on the same dataset at the same time (and thus the coordination through the spreadsheet)

[stephenbach]

@srush Does fine_grained_context_first render correctly as well? The first one looks same for me as your screenshot.

And yeah, +1, definitely a good motivating case for filters.

[shanyas10]

A bug I found (essentially a streamlit issue) was that the selected dataset (in the dropbox) changes as soon as we add a new template. This is because the rendered string in the dropdown changes ( count in parans increase by 1) and streamlit has some issues rendering such dynamic data (similar to this). I was also not able to find any workaround to fix this.

(Apologies if this has already been reported)

[srush]

@stephenbach fixed the formatting

@shanyas10 oh you are right! bummer. turned that off.

[VictorSanh]

@srush as discussed yesterday can you remove the "Template Viewer Section" as it duplicated from the Prompted dataset viwer?
I'll merge this pull requests then!

(I'll work on merging the 3 modes - sourcing, helicopter view, prompted dataset viewer - into a single app)

[srush]

@VictorSanh fixed.

[VictorSanh]

awesome, thank you, merging it!