JanekDev
commented
2 weeks ago - tested the model loading for forecasting, classification, imputation and representation learning
- because the dropout is not set in the HF config, the naming correction did not break anything
- all the dropouts are set to the default values specified in the model implementation, no dropout is set from the config directly (besides the t5 one, but it is well-separated inside the t5 config)