This PR fixes a compatibility issue introduced in ST 3.1 in this commit.
Basically, this added the possibility add module_kwargs in the config so that the model can leverage them for a specific processing, such as the dataset_embeddings of cde.
This fix allows to bump the version of ST to newer ones and also to handle such custom processing aswell (as we are extending the ST model).
raphaelsty
commented
1 month ago
This PR fixes a compatibility issue introduced in ST 3.1 in this commit.
Basically, this added the possibility add module_kwargs in the config so that the model can leverage them for a specific processing, such as the dataset_embeddings of cde.
This fix allows to bump the version of ST to newer ones and also to handle such custom processing aswell (as we are extending the ST model).
Edit for more information: the load_sbert_model function set the self.modules_kwargs that are then used in the forward pass of the Transformer model if some are set in the config.json
Related issue: https://github.com/lightonai/pylate/issues/64