Get the LLaMA backbone and its components closer to Mistral. Later, it would be nice to unify them to reduce code duplication. Tests haven't been modified, so the behavior of the components remain the same.
Note that I don't update the checkpoint file in this PR; it would be better to do it once the other components (preprocessor and generator) are in.
Get the LLaMA backbone and its components closer to Mistral. Later, it would be nice to unify them to reduce code duplication. Tests haven't been modified, so the behavior of the components remain the same.
Note that I don't update the checkpoint file in this PR; it would be better to do it once the other components (preprocessor and generator) are in.