Merge pull request #892 from modelscope/dev2-dzj

refine training framework

Author: Artiprocher

diffsynth/trainers/utils.py

@@ -1,4 +1,6 @@
 import imageio, os, torch, warnings, torchvision, argparse, json
+from ..utils import ModelConfig
+from ..models.utils import load_state_dict
 from peft import LoraConfig, inject_adapter_in_model
 from PIL import Image
 import pandas as pd
@@ -424,7 +426,53 @@ def transfer_data_to_device(self, data, device):
             if isinstance(data[key], torch.Tensor):
                 data[key] = data[key].to(device)
         return data
-
+    
+    
+    def parse_model_configs(self, model_paths, model_id_with_origin_paths, enable_fp8_training=False):
+        offload_dtype = torch.float8_e4m3fn if enable_fp8_training else None
+        model_configs = []
+        if model_paths is not None:
+            model_paths = json.loads(model_paths)
+            model_configs += [ModelConfig(path=path, offload_dtype=offload_dtype) for path in model_paths]
+        if model_id_with_origin_paths is not None:
+            model_id_with_origin_paths = model_id_with_origin_paths.split(",")
+            model_configs += [ModelConfig(model_id=i.split(":")[0], origin_file_pattern=i.split(":")[1], offload_dtype=offload_dtype) for i in model_id_with_origin_paths]
+        return model_configs
+    
+    
+    def switch_pipe_to_training_mode(
+        self,
+        pipe,
+        trainable_models,
+        lora_base_model, lora_target_modules, lora_rank, lora_checkpoint=None,
+        enable_fp8_training=False,
+    ):
+        # Scheduler
+        pipe.scheduler.set_timesteps(1000, training=True)
+        
+        # Freeze untrainable models
+        pipe.freeze_except([] if trainable_models is None else trainable_models.split(","))
+        
+        # Enable FP8 if pipeline supports
+        if enable_fp8_training and hasattr(pipe, "_enable_fp8_lora_training"):
+            pipe._enable_fp8_lora_training(torch.float8_e4m3fn)
+        
+        # Add LoRA to the base models
+        if lora_base_model is not None:
+            model = self.add_lora_to_model(
+                getattr(pipe, lora_base_model),
+                target_modules=lora_target_modules.split(","),
+                lora_rank=lora_rank,
+                upcast_dtype=pipe.torch_dtype,
+            )
+            if lora_checkpoint is not None:
+                state_dict = load_state_dict(lora_checkpoint)
+                state_dict = self.mapping_lora_state_dict(state_dict)
+                load_result = model.load_state_dict(state_dict, strict=False)
+                print(f"LoRA checkpoint loaded: {lora_checkpoint}, total {len(state_dict)} keys")
+                if len(load_result[1]) > 0:
+                    print(f"Warning, LoRA key mismatch! Unexpected keys in LoRA checkpoint: {load_result[1]}")
+            setattr(pipe, lora_base_model, model)
 
 
 class ModelLogger:
@@ -472,14 +520,26 @@ def launch_training_task(
     dataset: torch.utils.data.Dataset,
     model: DiffusionTrainingModule,
     model_logger: ModelLogger,
-    optimizer: torch.optim.Optimizer,
-    scheduler: torch.optim.lr_scheduler.LRScheduler,
+    learning_rate: float = 1e-5,
+    weight_decay: float = 1e-2,
     num_workers: int = 8,
     save_steps: int = None,
     num_epochs: int = 1,
     gradient_accumulation_steps: int = 1,
     find_unused_parameters: bool = False,
+    args = None,
 ):
+    if args is not None:
+        learning_rate = args.learning_rate
+        weight_decay = args.weight_decay
+        num_workers = args.dataset_num_workers
+        save_steps = args.save_steps
+        num_epochs = args.num_epochs
+        gradient_accumulation_steps = args.gradient_accumulation_steps
+        find_unused_parameters = args.find_unused_parameters
+    
+    optimizer = torch.optim.AdamW(model.trainable_modules(), lr=learning_rate, weight_decay=weight_decay)
+    scheduler = torch.optim.lr_scheduler.ConstantLR(optimizer)
     dataloader = torch.utils.data.DataLoader(dataset, shuffle=True, collate_fn=lambda x: x[0], num_workers=num_workers)
     accelerator = Accelerator(
         gradient_accumulation_steps=gradient_accumulation_steps,
@@ -509,8 +569,12 @@ def launch_data_process_task(
     model: DiffusionTrainingModule,
     model_logger: ModelLogger,
     num_workers: int = 8,
+    args = None,
 ):
-    dataloader = torch.utils.data.DataLoader(dataset, shuffle=True, collate_fn=lambda x: x[0], num_workers=num_workers)
+    if args is not None:
+        num_workers = args.dataset_num_workers
+        
+    dataloader = torch.utils.data.DataLoader(dataset, shuffle=False, collate_fn=lambda x: x[0], num_workers=num_workers)
     accelerator = Accelerator()
     model, dataloader = accelerator.prepare(model, dataloader)
 
@@ -520,7 +584,7 @@ def launch_data_process_task(
                 folder = os.path.join(model_logger.output_path, str(accelerator.process_index))
                 os.makedirs(folder, exist_ok=True)
                 save_path = os.path.join(model_logger.output_path, str(accelerator.process_index), f"{data_id}.pth")
-                data = model(data)
+                data = model(data, return_inputs=True)
                 torch.save(data, save_path)
 
 
@@ -623,4 +687,5 @@ def qwen_image_parser():
     parser.add_argument("--weight_decay", type=float, default=0.01, help="Weight decay.")
     parser.add_argument("--processor_path", type=str, default=None, help="Path to the processor. If provided, the processor will be used for image editing.")
     parser.add_argument("--enable_fp8_training", default=False, action="store_true", help="Whether to enable FP8 training. Only available for LoRA training on a single GPU.")
+    parser.add_argument("--task", type=str, default="sft", required=False, help="Task type.")
     return parser

examples/flux/model_training/train.py

@@ -20,37 +20,16 @@ def __init__(
     ):
         super().__init__()
         # Load models
-        model_configs = []
-        if model_paths is not None:
-            model_paths = json.loads(model_paths)
-            model_configs += [ModelConfig(path=path) for path in model_paths]
-        if model_id_with_origin_paths is not None:
-            model_id_with_origin_paths = model_id_with_origin_paths.split(",")
-            model_configs += [ModelConfig(model_id=i.split(":")[0], origin_file_pattern=i.split(":")[1]) for i in model_id_with_origin_paths]
+        model_configs = self.parse_model_configs(model_paths, model_id_with_origin_paths, enable_fp8_training=False)
         self.pipe = FluxImagePipeline.from_pretrained(torch_dtype=torch.bfloat16, device="cpu", model_configs=model_configs)
 
-        # Reset training scheduler
-        self.pipe.scheduler.set_timesteps(1000, training=True)
-        
-        # Freeze untrainable models
-        self.pipe.freeze_except([] if trainable_models is None else trainable_models.split(","))
+        # Training mode
+        self.switch_pipe_to_training_mode(
+            self.pipe, trainable_models,
+            lora_base_model, lora_target_modules, lora_rank, lora_checkpoint=lora_checkpoint,
+            enable_fp8_training=False,
+        )
 
-        # Add LoRA to the base models
-        if lora_base_model is not None:
-            model = self.add_lora_to_model(
-                getattr(self.pipe, lora_base_model),
-                target_modules=lora_target_modules.split(","),
-                lora_rank=lora_rank
-            )
-            if lora_checkpoint is not None:
-                state_dict = load_state_dict(lora_checkpoint)
-                state_dict = self.mapping_lora_state_dict(state_dict)
-                load_result = model.load_state_dict(state_dict, strict=False)
-                print(f"LoRA checkpoint loaded: {lora_checkpoint}, total {len(state_dict)} keys")
-                if len(load_result[1]) > 0:
-                    print(f"Warning, LoRA key mismatch! Unexpected keys in LoRA checkpoint: {load_result[1]}")
-            setattr(self.pipe, lora_base_model, model)
-            
         # Store other configs
         self.use_gradient_checkpointing = use_gradient_checkpointing
         self.use_gradient_checkpointing_offload = use_gradient_checkpointing_offload
@@ -138,13 +117,4 @@ def forward(self, data, inputs=None):
         remove_prefix_in_ckpt=args.remove_prefix_in_ckpt,
         state_dict_converter=FluxLoRAConverter.align_to_opensource_format if args.align_to_opensource_format else lambda x:x,
     )
-    optimizer = torch.optim.AdamW(model.trainable_modules(), lr=args.learning_rate, weight_decay=args.weight_decay)
-    scheduler = torch.optim.lr_scheduler.ConstantLR(optimizer)
-    launch_training_task(
-        dataset, model, model_logger, optimizer, scheduler,
-        num_epochs=args.num_epochs,
-        gradient_accumulation_steps=args.gradient_accumulation_steps,
-        save_steps=args.save_steps,
-        find_unused_parameters=args.find_unused_parameters,
-        num_workers=args.dataset_num_workers,
-    )
+    launch_training_task(dataset, model, model_logger, args=args)

examples/qwen_image/model_training/lora/Qwen-Image-Splited.sh

@@ -1,11 +1,12 @@
-accelerate launch examples/qwen_image/model_training/train_data_process.py \
+accelerate launch examples/qwen_image/model_training/train.py \
   --dataset_base_path data/example_image_dataset \
   --dataset_metadata_path data/example_image_dataset/metadata.csv \
   --max_pixels 1048576 \
   --model_id_with_origin_paths "Qwen/Qwen-Image:text_encoder/model*.safetensors,Qwen/Qwen-Image:vae/diffusion_pytorch_model.safetensors" \
   --output_path "./models/train/Qwen-Image_lora_cache" \
   --use_gradient_checkpointing \
-  --dataset_num_workers 8
+  --dataset_num_workers 8 \
+  --task data_process
 
 accelerate launch examples/qwen_image/model_training/train.py \
   --dataset_base_path models/train/Qwen-Image_lora_cache \

examples/qwen_image/model_training/train.py

@@ -2,7 +2,7 @@
 from diffsynth import load_state_dict
 from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
 from diffsynth.pipelines.flux_image_new import ControlNetInput
-from diffsynth.trainers.utils import DiffusionTrainingModule, ModelLogger, launch_training_task, qwen_image_parser
+from diffsynth.trainers.utils import DiffusionTrainingModule, ModelLogger, qwen_image_parser, launch_training_task, launch_data_process_task
 from diffsynth.trainers.unified_dataset import UnifiedDataset
 os.environ["TOKENIZERS_PARALLELISM"] = "false"
 
@@ -22,46 +22,18 @@ def __init__(
     ):
         super().__init__()
         # Load models
-        offload_dtype = torch.float8_e4m3fn if enable_fp8_training else None
-        model_configs = []
-        if model_paths is not None:
-            model_paths = json.loads(model_paths)
-            model_configs += [ModelConfig(path=path, offload_dtype=offload_dtype) for path in model_paths]
-        if model_id_with_origin_paths is not None:
-            model_id_with_origin_paths = model_id_with_origin_paths.split(",")
-            model_configs += [ModelConfig(model_id=i.split(":")[0], origin_file_pattern=i.split(":")[1], offload_dtype=offload_dtype) for i in model_id_with_origin_paths]
-
+        model_configs = self.parse_model_configs(model_paths, model_id_with_origin_paths, enable_fp8_training=enable_fp8_training)
         tokenizer_config = ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="tokenizer/") if tokenizer_path is None else ModelConfig(tokenizer_path)
         processor_config = ModelConfig(model_id="Qwen/Qwen-Image-Edit", origin_file_pattern="processor/") if processor_path is None else ModelConfig(processor_path)
         self.pipe = QwenImagePipeline.from_pretrained(torch_dtype=torch.bfloat16, device="cpu", model_configs=model_configs, tokenizer_config=tokenizer_config, processor_config=processor_config)
+
+        # Training mode
+        self.switch_pipe_to_training_mode(
+            self.pipe, trainable_models,
+            lora_base_model, lora_target_modules, lora_rank, lora_checkpoint=lora_checkpoint,
+            enable_fp8_training=enable_fp8_training,
+        )
 
-        # Enable FP8
-        if enable_fp8_training:
-            self.pipe._enable_fp8_lora_training(torch.float8_e4m3fn)
-        
-        # Reset training scheduler (do it in each training step)
-        self.pipe.scheduler.set_timesteps(1000, training=True)
-        
-        # Freeze untrainable models
-        self.pipe.freeze_except([] if trainable_models is None else trainable_models.split(","))
-        
-        # Add LoRA to the base models
-        if lora_base_model is not None:
-            model = self.add_lora_to_model(
-                getattr(self.pipe, lora_base_model),
-                target_modules=lora_target_modules.split(","),
-                lora_rank=lora_rank,
-                upcast_dtype=self.pipe.torch_dtype,
-            )
-            if lora_checkpoint is not None:
-                state_dict = load_state_dict(lora_checkpoint)
-                state_dict = self.mapping_lora_state_dict(state_dict)
-                load_result = model.load_state_dict(state_dict, strict=False)
-                print(f"LoRA checkpoint loaded: {lora_checkpoint}, total {len(state_dict)} keys")
-                if len(load_result[1]) > 0:
-                    print(f"Warning, LoRA key mismatch! Unexpected keys in LoRA checkpoint: {load_result[1]}")
-            setattr(self.pipe, lora_base_model, model)
-            
         # Store other configs
         self.use_gradient_checkpointing = use_gradient_checkpointing
         self.use_gradient_checkpointing_offload = use_gradient_checkpointing_offload
@@ -109,9 +81,10 @@ def forward_preprocess(self, data):
         return {**inputs_shared, **inputs_posi}
 
 
-    def forward(self, data, inputs=None):
+    def forward(self, data, inputs=None, return_inputs=False):
         if inputs is None: inputs = self.forward_preprocess(data)
         else: inputs = self.transfer_data_to_device(inputs, self.pipe.device)
+        if return_inputs: return inputs
         models = {name: getattr(self.pipe, name) for name in self.pipe.in_iteration_models}
         loss = self.pipe.training_loss(**models, **inputs)
         return loss
@@ -151,13 +124,8 @@ def forward(self, data, inputs=None):
         enable_fp8_training=args.enable_fp8_training,
     )
     model_logger = ModelLogger(args.output_path, remove_prefix_in_ckpt=args.remove_prefix_in_ckpt)
-    optimizer = torch.optim.AdamW(model.trainable_modules(), lr=args.learning_rate, weight_decay=args.weight_decay)
-    scheduler = torch.optim.lr_scheduler.ConstantLR(optimizer)
-    launch_training_task(
-        dataset, model, model_logger, optimizer, scheduler,
-        num_epochs=args.num_epochs,
-        gradient_accumulation_steps=args.gradient_accumulation_steps,
-        save_steps=args.save_steps,
-        find_unused_parameters=args.find_unused_parameters,
-        num_workers=args.dataset_num_workers,
-    )
+    launcher_map = {
+        "sft": launch_training_task,
+        "data_process": launch_data_process_task
+    }
+    launcher_map[args.task](dataset, model, model_logger, args=args)