fastvideo.v1.training.wan_training_pipeline

fastvideo.v1.training.wan_training_pipeline#

Module Contents#

Classes#

WanTrainingPipeline

A training pipeline for Wan.

Functions#

main

Data#

ENABLE_GRADIENT_CHECK

logger

vsa_available

API#

fastvideo.v1.training.wan_training_pipeline.ENABLE_GRADIENT_CHECK[source]#

False

class fastvideo.v1.training.wan_training_pipeline.WanTrainingPipeline(model_path: str, fastvideo_args: Union[fastvideo.v1.fastvideo_args.FastVideoArgs, fastvideo.v1.fastvideo_args.TrainingArgs], required_config_modules: Optional[List[str]] = None, loaded_modules: Optional[Dict[str, torch.nn.Module]] = None)[source]#

Bases: fastvideo.v1.training.training_pipeline.TrainingPipeline

A training pipeline for Wan.

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.v1.fastvideo_args.TrainingArgs)[source]#

May be used in future refactors.

forward(batch: fastvideo.v1.pipelines.pipeline_batch_info.ForwardBatch, fastvideo_args: fastvideo.v1.fastvideo_args.FastVideoArgs)[source]#
initialize_pipeline(fastvideo_args: fastvideo.v1.fastvideo_args.FastVideoArgs)[source]#
initialize_validation_pipeline(training_args: fastvideo.v1.fastvideo_args.TrainingArgs)[source]#
train_one_step(transformer, model_type, optimizer, lr_scheduler, loader_iter, noise_scheduler, noise_random_generator, gradient_accumulation_steps, sp_size, precondition_outputs, max_grad_norm, weighting_scheme, logit_mean, logit_std, mode_scale, patch_size, current_vsa_sparsity) → tuple[float, float][source]#
fastvideo.v1.training.wan_training_pipeline.logger[source]#

‘init_logger(…)’

fastvideo.v1.training.wan_training_pipeline.main(args) → None[source]#
fastvideo.v1.training.wan_training_pipeline.vsa_available[source]#

False

Contents