Created by: lilisierrayu
Patch Description "prefix_tokens" not used by sequence_generator or sequence_score, inference_step() implementation in lanague_modeling and language_modeling_inference_for_models_trained_with_streaming are the same as the one in base_task.
Testing steps