block wise controlnet

examples/qwen_image/model_training/full/Qwen-Image-BlockWiseControlNet.sh

@@ -0,0 +1,36 @@
+accelerate launch --config_file examples/qwen_image/model_training/full/accelerate_config.yaml examples/qwen_image/model_training/train.py \
+  --dataset_base_path "" \
+  --dataset_metadata_path data/t2i_dataset_annotations/blip3o/blip3o_control_images_train_for_diffsynth.jsonl \
+  --data_file_keys "image,controlnet_image" \
+  --max_pixels 1048576 \
+  --dataset_repeat 50 \
+  --model_paths '[
+    [
+        "models/Qwen/Qwen-Image/transformer/diffusion_pytorch_model-00001-of-00009.safetensors",
+        "models/Qwen/Qwen-Image/transformer/diffusion_pytorch_model-00002-of-00009.safetensors",
+        "models/Qwen/Qwen-Image/transformer/diffusion_pytorch_model-00003-of-00009.safetensors",
+        "models/Qwen/Qwen-Image/transformer/diffusion_pytorch_model-00004-of-00009.safetensors",
+        "models/Qwen/Qwen-Image/transformer/diffusion_pytorch_model-00005-of-00009.safetensors",
+        "models/Qwen/Qwen-Image/transformer/diffusion_pytorch_model-00006-of-00009.safetensors",
+        "models/Qwen/Qwen-Image/transformer/diffusion_pytorch_model-00007-of-00009.safetensors",
+        "models/Qwen/Qwen-Image/transformer/diffusion_pytorch_model-00008-of-00009.safetensors",
+        "models/Qwen/Qwen-Image/transformer/diffusion_pytorch_model-00009-of-00009.safetensors"
+    ],
+    [
+        "models/Qwen/Qwen-Image/text_encoder/model-00001-of-00004.safetensors",
+        "models/Qwen/Qwen-Image/text_encoder/model-00002-of-00004.safetensors",
+        "models/Qwen/Qwen-Image/text_encoder/model-00003-of-00004.safetensors",
+        "models/Qwen/Qwen-Image/text_encoder/model-00004-of-00004.safetensors"
+    ],
+    "models/Qwen/Qwen-Image/vae/diffusion_pytorch_model.safetensors",
+    "models/DiffSynth-Studio/BlockWiseControlnet/model_init.safetensors"
+]' \
+  --learning_rate 1e-3 \
+  --num_epochs 1000000 \
+  --remove_prefix_in_ckpt "pipe.blockwise_controlnet." \
+  --output_path "./models/train/Qwen-Image-BlockWiseControlNet_full_lr1e-3_wd1e-6" \
+  --trainable_models "blockwise_controlnet" \
+  --extra_inputs "controlnet_image" \
+  --use_gradient_checkpointing \
+  --dataset_num_workers 8 \
+  --save_steps 2000

examples/qwen_image/model_training/full/accelerate_config.yaml

@@ -0,0 +1,22 @@
+compute_environment: LOCAL_MACHINE
+debug: false
+deepspeed_config:
+  gradient_accumulation_steps: 1
+  offload_optimizer_device: none
+  offload_param_device: none
+  zero3_init_flag: false
+  zero_stage: 2
+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/qwen_image/model_training/full/others/initialize_blockwise_controlnet.py

@@ -0,0 +1,13 @@
+# This script is for initializing a Qwen-Image-ControlNet
+from diffsynth import load_state_dict, hash_state_dict_keys
+from diffsynth.models.qwen_image_controlnet import QwenImageBlockWiseControlNet
+import torch
+from safetensors.torch import save_file
+
+
+controlnet = QwenImageBlockWiseControlNet().to(dtype=torch.bfloat16, device="cuda")
+controlnet.init_weight()
+state_dict_controlnet = controlnet.state_dict()
+
+print(hash_state_dict_keys(state_dict_controlnet))
+save_file(state_dict_controlnet, "models/DiffSynth-Studio/BlockWiseControlnet/model_init.safetensors")

examples/qwen_image/model_training/train.py

@@ -118,7 +118,7 @@ if __name__ == "__main__":
         remove_prefix_in_ckpt=args.remove_prefix_in_ckpt,
         state_dict_converter=QwenImageLoRAConverter.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)
+    optimizer = torch.optim.AdamW(model.trainable_modules(), lr=args.learning_rate, weight_decay=0.000001)
     scheduler = torch.optim.lr_scheduler.ConstantLR(optimizer)
     launch_training_task(
         dataset, model, model_logger, optimizer, scheduler,

examples/qwen_image/model_training/validate_full/Qwen-Image-BlockWise-Controlnet.py

@@ -0,0 +1,38 @@
+from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
+from diffsynth.pipelines.flux_image_new import FluxImagePipeline, ModelConfig, ControlNetInput
+from diffsynth import load_state_dict
+import torch
+from PIL import Image
+from diffsynth.controlnets.processors import Annotator
+import os
+
+
+pipe = QwenImagePipeline.from_pretrained(
+    torch_dtype=torch.bfloat16,
+    device="cuda",
+    model_configs=[
+        ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="transformer/diffusion_pytorch_model*.safetensors"),
+        ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="text_encoder/model*.safetensors"),
+        ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="vae/diffusion_pytorch_model.safetensors"),
+        ModelConfig(path="models/DiffSynth-Studio/BlockWiseControlnet/model_init.safetensors"),
+    ],
+    tokenizer_config=ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="tokenizer/"),
+)
+
+state_dict = load_state_dict("models/train/Qwen-Image-BlockWiseControlNet_full_lr1e-3_wd1e-6/step-26000.safetensors")
+pipe.blockwise_controlnet.load_state_dict(state_dict)
+
+prompt = "精致肖像，水下少女，蓝裙飘逸，发丝轻扬，光影透澈，气泡环绕，面容恬静，细节精致，梦幻唯美。"
+image = Image.open("test_image.jpg").convert("RGB").resize((1024, 1024))
+canny_image = Annotator("canny")(image)
+canny_image.save("canny_image_test.jpg")
+
+controlnet_input = ControlNetInput(
+    image=canny_image,
+    scale=1.0,
+    processor_id="canny",
+)
+
+for seed in range(100, 200):
+    image = pipe(prompt, seed=seed, height=1024, width=1024, controlnet_inputs=[controlnet_input], num_inference_steps=30, cfg_scale=4.0)
+    image.save(f"test_image_controlnet_step2k_1_{seed}.jpg")