Training documentation

[ValentinJSchmidt]

Adds a little to the documentation and also puts the dropout to 0.7 in the Embedder Model so that Dropout is corrected in the .eval mode of the mode.
It is currently untested if this works

[Copilot]

Pull Request Overview

This PR aims to update the training documentation and correct the dropout behavior in the Embedder Model by setting the dropout rate to 0.7 during instantiation.

• Updates the dropout parameter when creating the EmbeddingModel instance
• Improves documentation regarding the dropout change in eval mode

Files not reviewed (3)

• README.rst: Language not supported
• docs/source/index.rst: Language not supported
• docs/source/training.rst: Language not supported

Comments suppressed due to low confidence (1)

paule/paule.py:165

• Consider adding test cases to verify that the new dropout parameter behaves as expected in eval mode.

self.embedder = EmbeddingModel(num_lstm_layers=2, hidden_size=720, dropout=0.7).double()