check if training entities in DIET is still working on e2e

[Ghostvv]

On e2e, we updated tf methods for TED, since these methods are reused in DIET, DIET performance on e2e branch should be benchmarked against DIET in master

[tttthomasssss]

@Ghostvv do you have any datasets in mind? Otherwise I would have gone for carbon-bot as a start, and then potentially some of the ones we've used in the DIET paper.

[Ghostvv]

@tttthomasssss I'd start by running all our datasets through GitHub as we do it for PRs

[tttthomasssss]

• touch #7289 #7371 is benchmarking DIET from master
• touch #7289 (e2e side) #7375 is benchmarking DIET from e2e

[tttthomasssss]

@Ghostvv results look pretty similar, the ones from e2e seem to be a tiny bit lower than from master (on carbon-bot and HERMIT), though I'd regard them within range still. The stuff from e2e also seems to be a bit slower (e.g. up to 2mins for carbon-bot and up to 5mins for HERMIT). But this might also due to github.

[Ghostvv]

Great! then I think we can close the issue, @tabergma do you have any other tests in mind?

[tabergma]

Great, no, I think that should be sufficient.