Training tables with more than 512 cells

google-research / tapas

End-to-end neural table-text understanding models.

Apache License 2.0

1.15k stars 217 forks source link

Training tables with more than 512 cells #158

Closed raunak-redseer closed 2 years ago

raunak-redseer commented 2 years ago

The original paper mentions that tables with maximum 500 cells were used to train the model. There is also a 512 token limit in the TAPAS config file, which implies that a table can have no more than 512 cells while predicting/inferencing. Just wanted to understand how to use TAPAS for large tables, for instance, for tables with ~100 Million rows and ~100 columns, if at all it is possible.

raunak-redseer commented 2 years ago

Closing it since similar issues had the supporting answers