transformers-interpret icon indicating copy to clipboard operation
transformers-interpret copied to clipboard

Support for Summarization models

Open jmdu99 opened this issue 2 years ago • 3 comments

Hi,

I would like to ask if this tool works with Seq2Seq models fine-tuned for summarization like BART or T5. Or if you plan to add this in the future. Thanks!

jmdu99 avatar Mar 09 '23 21:03 jmdu99

me too

datalee avatar Mar 10 '23 06:03 datalee

Me too plz! I already open an issue on Captum too. Thanks! Let me know if there is anything I can help we need this for a big project.

Fidac avatar Mar 22 '23 17:03 Fidac

I'm looking into this at the moment, it's kind of a tricky one to reason around because Seq2Seq models process one token at a time so the explanations would need to be calculated for each generated token with each new addition to the sequence accounted for, I'll give it a look and see if it can be done.

cdpierse avatar Apr 06 '23 23:04 cdpierse