How do you suggest we can further train the per-trainined ColBert model form this repo on our domain specific dataset.
Currently the repo only supports training the model from scratch on MS-MARCO dataset.
What should be the loss function for this, what should the config look like?
How do you suggest we can further train the per-trainined ColBert model form this repo on our domain specific dataset. Currently the repo only supports training the model from scratch on MS-MARCO dataset.
What should be the loss function for this, what should the config look like?