Support JoyAI-Image-Edit (#1393)

* auto intergrate joyimage model

* joyimage pipeline

* train

* ready

* styling

* joyai-image docs

* update readme

* pr review

examples/joyai_image/model_training/full/JoyAI-Image-Edit.sh

@@ -0,0 +1,35 @@
+# Dataset: data/diffsynth_example_dataset/joyai_image/JoyAI-Image-Edit/
+# Download: modelscope download --dataset DiffSynth-Studio/diffsynth_example_dataset --include "joyai_image/JoyAI-Image-Edit/*" --local_dir ./data/diffsynth_example_dataset
+
+accelerate launch examples/joyai_image/model_training/train.py \
+    --dataset_base_path "./data/diffsynth_example_dataset/joyai_image/JoyAI-Image-Edit" \
+    --dataset_metadata_path "./data/diffsynth_example_dataset/joyai_image/JoyAI-Image-Edit/metadata.csv" \
+    --max_pixels 1048576 \
+    --dataset_repeat 1 \
+    --model_id_with_origin_paths "jd-opensource/JoyAI-Image-Edit:JoyAI-Image-Und/model*.safetensors,jd-opensource/JoyAI-Image-Edit:vae/Wan2.1_VAE.pth" \
+    --learning_rate 1e-5 \
+    --num_epochs 2 \
+    --remove_prefix_in_ckpt "pipe.dit." \
+    --output_path "./models/train/JoyAI-Image-Edit-full-cache" \
+    --use_gradient_checkpointing \
+    --find_unused_parameters \
+    --data_file_keys "image,edit_image" \
+    --extra_inputs "edit_image" \
+    --task "sft:data_process"
+
+accelerate launch --config_file examples/joyai_image/model_training/full/accelerate_config_zero3.yaml \
+    examples/joyai_image/model_training/train.py \
+    --dataset_base_path "./models/train/JoyAI-Image-Edit-full-cache" \
+    --max_pixels 1048576 \
+    --dataset_repeat 50 \
+    --model_id_with_origin_paths "jd-opensource/JoyAI-Image-Edit:transformer/transformer.pth" \
+    --learning_rate 1e-5 \
+    --num_epochs 2 \
+    --remove_prefix_in_ckpt "pipe.dit." \
+    --output_path "./models/train/JoyAI-Image-Edit-full" \
+    --trainable_models "dit" \
+    --use_gradient_checkpointing \
+    --find_unused_parameters \
+    --data_file_keys "image,edit_image" \
+    --extra_inputs "edit_image" \
+    --task "sft:train"

examples/joyai_image/model_training/full/accelerate_config_zero3.yaml

@@ -0,0 +1,23 @@
+compute_environment: LOCAL_MACHINE
+debug: false
+deepspeed_config:
+  gradient_accumulation_steps: 1
+  offload_optimizer_device: none
+  offload_param_device: none
+  zero3_init_flag: true
+  zero3_save_16bit_model: true
+  zero_stage: 3
+distributed_type: DEEPSPEED
+downcast_bf16: 'no'
+enable_cpu_affinity: false
+machine_rank: 0
+main_training_function: main
+mixed_precision: bf16
+num_machines: 1
+num_processes: 8
+rdzv_backend: static
+same_network: true
+tpu_env: []
+tpu_use_cluster: false
+tpu_use_sudo: false
+use_cpu: false

examples/joyai_image/model_training/lora/JoyAI-Image-Edit.sh

@@ -0,0 +1,39 @@
+# Dataset: data/diffsynth_example_dataset/joyai_image/JoyAI-Image-Edit/
+# Download: modelscope download --dataset DiffSynth-Studio/diffsynth_example_dataset --include "joyai_image/JoyAI-Image-Edit/*" --local_dir ./data/diffsynth_example_dataset
+
+accelerate launch examples/joyai_image/model_training/train.py \
+    --dataset_base_path "./data/diffsynth_example_dataset/joyai_image/JoyAI-Image-Edit" \
+    --dataset_metadata_path "./data/diffsynth_example_dataset/joyai_image/JoyAI-Image-Edit/metadata.csv" \
+    --max_pixels 1048576 \
+    --dataset_repeat 1 \
+    --model_id_with_origin_paths "jd-opensource/JoyAI-Image-Edit:JoyAI-Image-Und/model*.safetensors,jd-opensource/JoyAI-Image-Edit:vae/Wan2.1_VAE.pth" \
+    --learning_rate 1e-4 \
+    --num_epochs 5 \
+    --remove_prefix_in_ckpt "pipe.dit." \
+    --output_path "./models/train/JoyAI-Image-Edit-split-cache" \
+    --lora_base_model "dit" \
+    --lora_target_modules "img_attn_qkv,txt_attn_qkv,img_attn_proj,txt_attn_proj" \
+    --lora_rank 32 \
+    --use_gradient_checkpointing \
+    --find_unused_parameters \
+    --data_file_keys "image,edit_image" \
+    --extra_inputs "edit_image" \
+    --task "sft:data_process"
+
+accelerate launch examples/joyai_image/model_training/train.py \
+    --dataset_base_path "./models/train/JoyAI-Image-Edit-split-cache" \
+    --max_pixels 1048576 \
+    --dataset_repeat 50 \
+    --model_id_with_origin_paths "jd-opensource/JoyAI-Image-Edit:transformer/transformer.pth" \
+    --learning_rate 1e-4 \
+    --num_epochs 5 \
+    --remove_prefix_in_ckpt "pipe.dit." \
+    --output_path "./models/train/JoyAI-Image-Edit-lora" \
+    --lora_base_model "dit" \
+    --lora_target_modules "img_attn_qkv,txt_attn_qkv,img_attn_proj,txt_attn_proj" \
+    --lora_rank 32 \
+    --use_gradient_checkpointing \
+    --find_unused_parameters \
+    --data_file_keys "image,edit_image" \
+    --extra_inputs "edit_image" \
+    --task "sft:train"

examples/joyai_image/model_training/train.py

@@ -0,0 +1,138 @@
+import torch, os, argparse, accelerate
+from diffsynth.core import UnifiedDataset
+from diffsynth.pipelines.joyai_image import JoyAIImagePipeline, ModelConfig
+from diffsynth.diffusion import *
+from diffsynth.core.data.operators import *
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+
+
+class JoyAIImageTrainingModule(DiffusionTrainingModule):
+    def __init__(
+        self,
+        model_paths=None, model_id_with_origin_paths=None,
+        processor_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,
+        device="cpu",
+        task="sft",
+    ):
+        super().__init__()
+        # Load models
+        model_configs = self.parse_model_configs(model_paths, model_id_with_origin_paths, fp8_models=fp8_models, offload_models=offload_models, device=device)
+        processor_config = ModelConfig(model_id="jd-opensource/JoyAI-Image-Edit", origin_file_pattern="JoyAI-Image-Und/") if processor_path is None else ModelConfig(processor_path)
+        self.pipe = JoyAIImagePipeline.from_pretrained(torch_dtype=torch.bfloat16, device=device, model_configs=model_configs, processor_config=processor_config)
+        self.pipe = self.split_pipeline_units(task, self.pipe, trainable_models, lora_base_model)
+
+        # Training mode
+        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,
+        )
+
+        # Other configs
+        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"]}
+        inputs_nega = {"negative_prompt": ""}
+        inputs_shared = {
+            # Assume you are using this pipeline for inference,
+            # please fill in the input parameters.
+            "input_image": data["image"],
+            "height": data["image"].size[1],
+            "width": data["image"].size[0],
+            # Please do not modify the following parameters
+            # unless you clearly know what this will cause.
+            "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 joyai_image_parser():
+    parser = argparse.ArgumentParser(description="JoyAI-Image training.")
+    parser = add_general_config(parser)
+    parser = add_image_size_config(parser)
+    parser.add_argument("--processor_path", type=str, default=None, help="Path to the processor.")
+    parser.add_argument("--initialize_model_on_cpu", default=False, action="store_true", help="Whether to initialize models on CPU.")
+    return parser
+
+
+if __name__ == "__main__":
+    parser = joyai_image_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=UnifiedDataset.default_image_operator(
+            base_path=args.dataset_base_path,
+            max_pixels=args.max_pixels,
+            height=args.height,
+            width=args.width,
+            height_division_factor=16,
+            width_division_factor=16,
+        ),
+    )
+    model = JoyAIImageTrainingModule(
+        model_paths=args.model_paths,
+        model_id_with_origin_paths=args.model_id_with_origin_paths,
+        processor_path=args.processor_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,
+        task=args.task,
+        device="cpu" if args.initialize_model_on_cpu else accelerator.device,
+    )
+    model_logger = ModelLogger(
+        args.output_path,
+        remove_prefix_in_ckpt=args.remove_prefix_in_ckpt,
+    )
+    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)

examples/joyai_image/model_training/validate_full/JoyAI-Image-Edit.py

@@ -0,0 +1,32 @@
+import torch
+from PIL import Image
+from diffsynth.pipelines.joyai_image import JoyAIImagePipeline, ModelConfig
+from diffsynth import load_state_dict
+
+pipe = JoyAIImagePipeline.from_pretrained(
+    torch_dtype=torch.bfloat16,
+    device="cuda",
+    model_configs=[
+        ModelConfig(model_id="jd-opensource/JoyAI-Image-Edit", origin_file_pattern="transformer/transformer.pth"),
+        ModelConfig(model_id="jd-opensource/JoyAI-Image-Edit", origin_file_pattern="JoyAI-Image-Und/model*.safetensors"),
+        ModelConfig(model_id="jd-opensource/JoyAI-Image-Edit", origin_file_pattern="vae/Wan2.1_VAE.pth"),
+    ],
+    processor_config=ModelConfig(model_id="jd-opensource/JoyAI-Image-Edit", origin_file_pattern="JoyAI-Image-Und/"),
+)
+
+state_dict = load_state_dict("models/train/JoyAI-Image-Edit_full/epoch-1.safetensors")
+pipe.dit.load_state_dict(state_dict)
+
+prompt = "将裙子改为粉色"
+edit_image = Image.open("data/diffsynth_example_dataset/joyai_image/JoyAI-Image-Edit/edit/image1.jpg").convert("RGB")
+
+image = pipe(
+    prompt=prompt,
+    edit_image=edit_image,
+    height=1024,
+    width=1024,
+    seed=0,
+    num_inference_steps=50,
+    cfg_scale=5.0,
+)
+image.save("image_full.jpg")

examples/joyai_image/model_training/validate_lora/JoyAI-Image-Edit.py

@@ -0,0 +1,30 @@
+import torch
+from PIL import Image
+from diffsynth.pipelines.joyai_image import JoyAIImagePipeline, ModelConfig
+
+pipe = JoyAIImagePipeline.from_pretrained(
+    torch_dtype=torch.bfloat16,
+    device="cuda",
+    model_configs=[
+        ModelConfig(model_id="jd-opensource/JoyAI-Image-Edit", origin_file_pattern="transformer/transformer.pth"),
+        ModelConfig(model_id="jd-opensource/JoyAI-Image-Edit", origin_file_pattern="JoyAI-Image-Und/model*.safetensors"),
+        ModelConfig(model_id="jd-opensource/JoyAI-Image-Edit", origin_file_pattern="vae/Wan2.1_VAE.pth"),
+    ],
+    processor_config=ModelConfig(model_id="jd-opensource/JoyAI-Image-Edit", origin_file_pattern="JoyAI-Image-Und/"),
+)
+
+pipe.load_lora(pipe.dit, "models/train/JoyAI-Image-Edit-lora/epoch-4.safetensors")
+
+prompt = "将裙子改为粉色"
+edit_image = Image.open("data/diffsynth_example_dataset/joyai_image/JoyAI-Image-Edit/edit/image1.jpg").convert("RGB")
+
+image = pipe(
+    prompt=prompt,
+    edit_image=edit_image,
+    height=1024,
+    width=1024,
+    seed=0,
+    num_inference_steps=30,
+    cfg_scale=5.0,
+)
+image.save("image_lora.jpg")