search

UKPLab / sentence-transformers

State-of-the-Art Text Embeddings
https://www.sbert.net
Apache License 2.0
14.78k stars 2.43k forks source link

Adding support for tabular data using Tapas introduced in transformer 4.1 #644

Open rostandkenne opened 3 years ago

rostandkenne commented 3 years ago

I must start by saying that is an awesome library that even simplify training cross-encoder.

Quick question, are there any plan to add support for tabular data in cross-encoder ? This recently became available in Huggingface transformer (tapasforsequenceclassification).

https://huggingface.co/transformers/master/model_doc/tapas.html#tapasforsequenceclassification

nreimers commented 3 years ago

Hi @joeyfromnijmegen The biggest issue is the different input format. This would require quite large changes in the cross-encoder class to enable the input of table data. So I am afraid that it will not be available soon.

But if you like to work on, I am happy about a Pull Requests which integrates Tapas into the cross-encoder wrapper.