fastvideo.training.wan_i2v_distillation_pipeline
#
Module Contents#
Classes#
A distillation pipeline for Wan that uses a single transformer model. The main transformer serves as the student model, and copies are made for teacher and critic. |
Functions#
Data#
API#
- class fastvideo.training.wan_i2v_distillation_pipeline.WanI2VDistillationPipeline(model_path: str, fastvideo_args: fastvideo.fastvideo_args.TrainingArgs, required_config_modules: list[str] | None = None, loaded_modules: dict[str, torch.nn.Module] | None = None)[source]#
Bases:
fastvideo.training.distillation_pipeline.DistillationPipeline
A distillation pipeline for Wan that uses a single transformer model. The main transformer serves as the student model, and copies are made for teacher and critic.
Initialization
Initialize the pipeline. After init, the pipeline should be ready to use. The pipeline should be stateless and not hold any batch state.
- create_training_stages(training_args: fastvideo.fastvideo_args.TrainingArgs)[source]#
May be used in future refactors.
- initialize_pipeline(fastvideo_args: fastvideo.fastvideo_args.FastVideoArgs)[source]#
Initialize Wan-specific scheduler.
- initialize_validation_pipeline(training_args: fastvideo.fastvideo_args.TrainingArgs)[source]#