This PR (if it's been done right), should not break any existing functionality. It's adds a lot a new aliases for existing symbols, that going forward will be the main way we document them.
Classifier -> TextClassifier and XXClassifer -> XXTextClassifier.
Add new base classes keras_nlp.models.TextClassifierPrepreprocessor, keras_nlp.models.CausalLMPreprocessor and keras_nlp.models.MaskedLMPreprocessor.
There's no need to update if you don't want to, we will continue to to support the existing usages indefinitely. This is primarily to reduce confusion in our own docs; we want ImageClasssifier and TextClassifier to have obviously distinct usage.
This new base classes also enable a cross-model way of writing task code with split preprocessing.
preprocessor = keras_nlp.models.TextClassifierPrepreprocessor.from_preset(
"bert_base_en",
)
classifier = keras_nlp.models.TextClassifier(
"bert_base_en",
num_classes=2,
)
... run prepreprocessing and training separately
As a follow up, I think we can remove a lot of code by pushing common code onto base classes. But I will do that in a later PR to keep this from getting to big.
This PR (if it's been done right), should not break any existing functionality. It's adds a lot a new aliases for existing symbols, that going forward will be the main way we document them.
Classifier->TextClassifierandXXClassifer->XXTextClassifier.keras_nlp.models.TextClassifierPrepreprocessor,keras_nlp.models.CausalLMPreprocessorandkeras_nlp.models.MaskedLMPreprocessor.There's no need to update if you don't want to, we will continue to to support the existing usages indefinitely. This is primarily to reduce confusion in our own docs; we want
ImageClasssifierandTextClassifierto have obviously distinct usage.This new base classes also enable a cross-model way of writing task code with split preprocessing.
As a follow up, I think we can remove a lot of code by pushing common code onto base classes. But I will do that in a later PR to keep this from getting to big.