search

terrierteam / pyterrier_colbert

82 stars 35 forks source link

Number of epochs to train the Colbert model in the fit( ) #37

Closed poorya2021 closed 2 years ago

poorya2021 commented 2 years ago

Hi everyone I had a question about training the Colbert model

How can I change the epoch and batch size for the fit( ) function used to train the model?

What is the number of default epochs in the fit( ) function?

Which part can I see the settings for the number of epochs and batch sizes for the fit( ) function?

Thank you

cmacdonald commented 2 years ago

pyterrier_colbert doesnt expose a fit function. The ColBERT documentation on training applies.

poorya2021 commented 2 years ago

Then @cmacdonald Is there no function or method for training Pyterrier_ColBERT in the Pyterrier package?

cmacdonald commented 2 years ago

The model that we provide in our notebooks is trained on MSMARCO passage ranking, and works effectively on many other corpora in a zero-shot fashion. To build that model, we followed the training instructions at https://github.com/stanford-futuredata/ColBERT#training and trained for 44,500 steps.

Training as part of a PyTerrier pipeline would be more expensive, as ranking would need to be performed, etc.

poorya2021 commented 2 years ago

Ok thank you