examples/ace_step/model_training/train.py

import os
import torch
import math
import argparse
import accelerate
from diffsynth.core import UnifiedDataset
from diffsynth.core.data.operators import ToAbsolutePath, LoadPureAudioWithTorchaudio
from diffsynth.pipelines.ace_step import AceStepPipeline, ModelConfig
from diffsynth.diffusion import *
os.environ["TOKENIZERS_PARALLELISM"] = "false"


class AceStepTrainingModule(DiffusionTrainingModule):
    def __init__(
        self,
        model_paths=None, model_id_with_origin_paths=None,
        tokenizer_path=None, silence_latent_path=None,
        trainable_models=None,
        lora_base_model=None, lora_target_modules="", lora_rank=32, lora_checkpoint=None,
        preset_lora_path=None, preset_lora_model=None,
        use_gradient_checkpointing=True,
        use_gradient_checkpointing_offload=False,
        extra_inputs=None,
        fp8_models=None,
        offload_models=None,
        quant_options=None,
        template_model_id_or_path=None,
        resume_from_checkpoint=None, remove_prefix_in_ckpt=None,
        device="cpu",
        task="sft",
    ):
        super().__init__()
        model_configs = self.parse_model_configs(model_paths, model_id_with_origin_paths, fp8_models=fp8_models, offload_models=offload_models, quant_options=quant_options, device=device)
        text_tokenizer_config = self.parse_path_or_model_id(tokenizer_path, default_value=ModelConfig(model_id="ACE-Step/Ace-Step1.5", origin_file_pattern="Qwen3-Embedding-0.6B/"))
        silence_latent_config = self.parse_path_or_model_id(silence_latent_path, default_value=ModelConfig(model_id="ACE-Step/Ace-Step1.5", origin_file_pattern="acestep-v15-turbo/silence_latent.pt"))
        self.pipe = AceStepPipeline.from_pretrained(
            torch_dtype=torch.bfloat16, device=device, model_configs=model_configs,
            text_tokenizer_config=text_tokenizer_config, silence_latent_config=silence_latent_config,
        )
        self.pipe = self.load_training_template_model(self.pipe, template_model_id_or_path, use_gradient_checkpointing, use_gradient_checkpointing_offload)
        self.pipe = self.split_pipeline_units(task, self.pipe, trainable_models, lora_base_model)
        self.resume_from_checkpoint(resume_from_checkpoint, remove_prefix_in_ckpt)

        self.switch_pipe_to_training_mode(
            self.pipe, trainable_models,
            lora_base_model, lora_target_modules, lora_rank, lora_checkpoint,
            preset_lora_path, preset_lora_model,
            task=task,
        )

        self.use_gradient_checkpointing = use_gradient_checkpointing
        self.use_gradient_checkpointing_offload = use_gradient_checkpointing_offload
        self.extra_inputs = extra_inputs.split(",") if extra_inputs is not None else []
        self.fp8_models = fp8_models
        self.task = task
        self.task_to_loss = {
            "sft:data_process": lambda pipe, *args: args,
            "sft": lambda pipe, inputs_shared, inputs_posi, inputs_nega: FlowMatchSFTLoss(pipe, **inputs_shared, **inputs_posi),
            "sft:train": lambda pipe, inputs_shared, inputs_posi, inputs_nega: FlowMatchSFTLoss(pipe, **inputs_shared, **inputs_posi),
        }

    def get_pipeline_inputs(self, data):
        inputs_posi = {"prompt": data["prompt"], "positive": True}
        inputs_nega = {"positive": False}
        duration = math.floor(data['audio'][0].shape[1] / data['audio'][1]) if data.get("audio") is not None else data.get("duration", 60)
        inputs_shared = {
            "input_audio": data["audio"],
            "lyrics": data["lyrics"],
            "task_type": "text2music",
            "duration": duration,
            "bpm": data.get("bpm", 100),
            "keyscale": data.get("keyscale", "C major"),
            "timesignature": data.get("timesignature", "4"),
            "vocal_language": data.get("vocal_language", "unknown"),
            "cfg_scale": 1,
            "rand_device": self.pipe.device,
            "use_gradient_checkpointing": self.use_gradient_checkpointing,
            "use_gradient_checkpointing_offload": self.use_gradient_checkpointing_offload,
        }
        inputs_shared = self.parse_extra_inputs(data, self.extra_inputs, inputs_shared)
        return inputs_shared, inputs_posi, inputs_nega

    def forward(self, data, inputs=None):
        if inputs is None: inputs = self.get_pipeline_inputs(data)
        inputs = self.transfer_data_to_device(inputs, self.pipe.device, self.pipe.torch_dtype)
        for unit in self.pipe.units:
            inputs = self.pipe.unit_runner(unit, self.pipe, *inputs)
        loss = self.task_to_loss[self.task](self.pipe, *inputs)
        return loss


def ace_step_parser():
    parser = argparse.ArgumentParser(description="ACE-Step training.")
    parser = add_general_config(parser)
    parser.add_argument("--tokenizer_path", type=str, default=None, help="Tokenizer path in format model_id:origin_pattern.")
    parser.add_argument("--silence_latent_path", type=str, default=None, help="Silence latent path in format model_id:origin_pattern.")
    parser.add_argument("--initialize_model_on_cpu", default=False, action="store_true", help="Whether to initialize models on CPU.")
    parser.add_argument("--max_audio_duration", type=int, default=None, help="Maximum audio length. Audio exceeding the length limit will be truncated.")
    return parser


if __name__ == "__main__":
    parser = ace_step_parser()
    args = parser.parse_args()
    accelerator = accelerate.Accelerator(
        gradient_accumulation_steps=args.gradient_accumulation_steps,
        kwargs_handlers=[accelerate.DistributedDataParallelKwargs(find_unused_parameters=args.find_unused_parameters)],
    )
    dataset = UnifiedDataset(
        base_path=args.dataset_base_path,
        metadata_path=args.dataset_metadata_path,
        repeat=args.dataset_repeat,
        data_file_keys=args.data_file_keys.split(","),
        main_data_operator=ToAbsolutePath(args.dataset_base_path) >> LoadPureAudioWithTorchaudio(target_sample_rate=48000, max_audio_duration=args.max_audio_duration),
    )
    model = AceStepTrainingModule(
        model_paths=args.model_paths,
        model_id_with_origin_paths=args.model_id_with_origin_paths,
        tokenizer_path=args.tokenizer_path,
        silence_latent_path=args.silence_latent_path,
        trainable_models=args.trainable_models,
        lora_base_model=args.lora_base_model,
        lora_target_modules=args.lora_target_modules,
        lora_rank=args.lora_rank,
        lora_checkpoint=args.lora_checkpoint,
        preset_lora_path=args.preset_lora_path,
        preset_lora_model=args.preset_lora_model,
        use_gradient_checkpointing=args.use_gradient_checkpointing,
        use_gradient_checkpointing_offload=args.use_gradient_checkpointing_offload,
        extra_inputs=args.extra_inputs,
        fp8_models=args.fp8_models,
        offload_models=args.offload_models,
        quant_options=args.quant_options,
        template_model_id_or_path=args.template_model_id_or_path,
        resume_from_checkpoint=args.resume_from_checkpoint,
        remove_prefix_in_ckpt=args.remove_prefix_in_ckpt,
        task=args.task,
        device="cpu" if (args.initialize_model_on_cpu or args.enable_model_cpu_offload) else accelerator.device,
    )
    model_logger = ModelLogger(
        args.output_path,
        remove_prefix_in_ckpt=args.remove_prefix_in_ckpt,
        enable_tensorboard_log=args.enable_tensorboard_log,
        enable_swanlab_log=args.enable_swanlab_log,
        swanlab_project=args.swanlab_project,
        enable_wandb_log=args.enable_wandb_log,
        wandb_project=args.wandb_project,
        enable_csv_log=args.enable_csv_log,
    )
    launcher_map = {
        "sft:data_process": launch_data_process_task,
        "sft": launch_training_task,
        "sft:train": launch_training_task,
    }
    launcher_map[args.task](accelerator, dataset, model, model_logger, args=args)