Default model no longer seems coherent

[indicodatasolutions]

Default model no longer seems coherent

[madisonmay]

Hi @bgreen-litl, thanks for the sanity check, sorry I didn't notice this sooner.

The 0.5.0 release involved a major refactor to migrate to tensorflow's estimator framework. It's quite possible that during this process, the code for generate_text was migrated improperly.

However, we ran a set of internal benchmarks to confirm that we could replicate previous model finetuning results on a set of a dozen datasets, so I think you'll find that if that is what you're using finetune for things will still function as intended.

Will try to get a patch for the generate_text() issue up as soon as possible! Thanks again for the bug report.

--Madison

[benleetownsend]

Apologies for this, I have fixed this in #190. The following lines show an example output.

from finetune import Classifier
a = Classifier()
INFO:finetune: Visible Devices: {1: GeForce GTX 1080 Ti}
a.generate_text()
'start\n " i do n't know , " i said . " i think i 'm going to go back to the hotel and get some sleep . " \n " you 're not going to sleep ? ".....