diff --git a/README.md b/README.md index d014bac..1acdff9 100644 --- a/README.md +++ b/README.md @@ -97,7 +97,7 @@ image.save("image.jpg") |[Qwen/Qwen-Image-Edit](https://www.modelscope.cn/models/Qwen/Qwen-Image-Edit)|[code](./examples/qwen_image/model_inference/Qwen-Image-Edit.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Edit.py)|[code](./examples/qwen_image/model_training/full/Qwen-Image-Edit.sh)|[code](./examples/qwen_image/model_training/validate_full/Qwen-Image-Edit.py)|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Edit.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Edit.py)| |[DiffSynth-Studio/Qwen-Image-EliGen-V2](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen-V2)|[code](./examples/qwen_image/model_inference/Qwen-Image-EliGen-V2.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-EliGen-V2.py)|-|-|[code](./examples/qwen_image/model_training/lora/Qwen-Image-EliGen.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-EliGen.py)| |[DiffSynth-Studio/Qwen-Image-Distill-Full](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-Full)|[code](./examples/qwen_image/model_inference/Qwen-Image-Distill-Full.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Distill-Full.py)|[code](./examples/qwen_image/model_training/full/Qwen-Image-Distill-Full.sh)|[code](./examples/qwen_image/model_training/validate_full/Qwen-Image-Distill-Full.py)|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Distill-Full.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Distill-Full.py)| -|[DiffSynth-Studio/Qwen-Image-Distill-LoRA](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-LoRA)|[code](./examples/qwen_image/model_inference/Qwen-Image-Distill-LoRA.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Distill-LoRA.py)|-|-|-|-| +|[DiffSynth-Studio/Qwen-Image-Distill-LoRA](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-LoRA)|[code](./examples/qwen_image/model_inference/Qwen-Image-Distill-LoRA.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Distill-LoRA.py)|-|-|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Distill-LoRA.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Distill-LoRA.py)| |[DiffSynth-Studio/Qwen-Image-EliGen](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen)|[code](./examples/qwen_image/model_inference/Qwen-Image-EliGen.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-EliGen.py)|-|-|[code](./examples/qwen_image/model_training/lora/Qwen-Image-EliGen.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-EliGen.py)| |[DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Canny](https://modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Canny)|[code](./examples/qwen_image/model_inference/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./examples/qwen_image/model_training/full/Qwen-Image-Blockwise-ControlNet-Canny.sh)|[code](./examples/qwen_image/model_training/validate_full/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Blockwise-ControlNet-Canny.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Blockwise-ControlNet-Canny.py)| |[DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Depth](https://modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Depth)|[code](./examples/qwen_image/model_inference/Qwen-Image-Blockwise-ControlNet-Depth.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Blockwise-ControlNet-Depth.py)|[code](./examples/qwen_image/model_training/full/Qwen-Image-Blockwise-ControlNet-Depth.sh)|[code](./examples/qwen_image/model_training/validate_full/Qwen-Image-Blockwise-ControlNet-Depth.py)|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Blockwise-ControlNet-Depth.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Blockwise-ControlNet-Depth.py)| @@ -380,6 +380,9 @@ https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/59fb2f7b-8de0-44 ## Update History + +- **September 9, 2025**: Our training framework now supports multiple training modes and has been adapted for Qwen-Image. In addition to the standard SFT training mode, Direct Distill is now also supported; please refer to [our example code](./examples/qwen_image/model_training/lora/Qwen-Image-Distill-LoRA.sh). This feature is experimental, and we will continue to improve it to support comprehensive model training capabilities. + - **August 28, 2025** We support Wan2.2-S2V, an audio-driven cinematic video generation model open-sourced by Alibaba. See [./examples/wanvideo/](./examples/wanvideo/). - **August 21, 2025**: [DiffSynth-Studio/Qwen-Image-EliGen-V2](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen-V2) is released! Compared to the V1 version, the training dataset has been updated to the [Qwen-Image-Self-Generated-Dataset](https://www.modelscope.cn/datasets/DiffSynth-Studio/Qwen-Image-Self-Generated-Dataset), enabling generated images to better align with the inherent image distribution and style of Qwen-Image. Please refer to [our sample code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-EliGen-V2.py). diff --git a/README_zh.md b/README_zh.md index edba5ab..bba1659 100644 --- a/README_zh.md +++ b/README_zh.md @@ -99,7 +99,7 @@ image.save("image.jpg") |[Qwen/Qwen-Image-Edit](https://www.modelscope.cn/models/Qwen/Qwen-Image-Edit)|[code](./examples/qwen_image/model_inference/Qwen-Image-Edit.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Edit.py)|[code](./examples/qwen_image/model_training/full/Qwen-Image-Edit.sh)|[code](./examples/qwen_image/model_training/validate_full/Qwen-Image-Edit.py)|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Edit.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Edit.py)| |[DiffSynth-Studio/Qwen-Image-EliGen-V2](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen-V2)|[code](./examples/qwen_image/model_inference/Qwen-Image-EliGen-V2.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-EliGen-V2.py)|-|-|[code](./examples/qwen_image/model_training/lora/Qwen-Image-EliGen.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-EliGen.py)| |[DiffSynth-Studio/Qwen-Image-Distill-Full](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-Full)|[code](./examples/qwen_image/model_inference/Qwen-Image-Distill-Full.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Distill-Full.py)|[code](./examples/qwen_image/model_training/full/Qwen-Image-Distill-Full.sh)|[code](./examples/qwen_image/model_training/validate_full/Qwen-Image-Distill-Full.py)|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Distill-Full.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Distill-Full.py)| -|[DiffSynth-Studio/Qwen-Image-Distill-LoRA](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-LoRA)|[code](./examples/qwen_image/model_inference/Qwen-Image-Distill-LoRA.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Distill-LoRA.py)|-|-|-|-| +|[DiffSynth-Studio/Qwen-Image-Distill-LoRA](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-LoRA)|[code](./examples/qwen_image/model_inference/Qwen-Image-Distill-LoRA.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Distill-LoRA.py)|-|-|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Distill-LoRA.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Distill-LoRA.py)| |[DiffSynth-Studio/Qwen-Image-EliGen](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen)|[code](./examples/qwen_image/model_inference/Qwen-Image-EliGen.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-EliGen.py)|-|-|[code](./examples/qwen_image/model_training/lora/Qwen-Image-EliGen.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-EliGen.py)| |[DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Canny](https://modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Canny)|[code](./examples/qwen_image/model_inference/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./examples/qwen_image/model_training/full/Qwen-Image-Blockwise-ControlNet-Canny.sh)|[code](./examples/qwen_image/model_training/validate_full/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Blockwise-ControlNet-Canny.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Blockwise-ControlNet-Canny.py)| |[DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Depth](https://modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Depth)|[code](./examples/qwen_image/model_inference/Qwen-Image-Blockwise-ControlNet-Depth.py)|[code](./examples/qwen_image/model_inference_low_vram/Qwen-Image-Blockwise-ControlNet-Depth.py)|[code](./examples/qwen_image/model_training/full/Qwen-Image-Blockwise-ControlNet-Depth.sh)|[code](./examples/qwen_image/model_training/validate_full/Qwen-Image-Blockwise-ControlNet-Depth.py)|[code](./examples/qwen_image/model_training/lora/Qwen-Image-Blockwise-ControlNet-Depth.sh)|[code](./examples/qwen_image/model_training/validate_lora/Qwen-Image-Blockwise-ControlNet-Depth.py)| @@ -396,6 +396,9 @@ https://github.com/Artiprocher/DiffSynth-Studio/assets/35051019/59fb2f7b-8de0-44 ## 更新历史 + +- **2025年9月9日** 我们的训练框架支持了多种训练模式，目前已适配 Qwen-Image，除标准 SFT 训练模式外，已支持 Direct Distill，请参考[我们的示例代码](./examples/qwen_image/model_training/lora/Qwen-Image-Distill-LoRA.sh)。这项功能是实验性的，我们将会继续完善已支持更全面的模型训练功能。 + - **2025年8月28日** 我们支持了Wan2.2-S2V，一个音频驱动的电影级视频生成模型。请参见[./examples/wanvideo/](./examples/wanvideo/)。 - **2025年8月21日** [DiffSynth-Studio/Qwen-Image-EliGen-V2](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen-V2) 发布！相比于 V1 版本，训练数据集变为 [Qwen-Image-Self-Generated-Dataset](https://www.modelscope.cn/datasets/DiffSynth-Studio/Qwen-Image-Self-Generated-Dataset)，因此，生成的图像更符合 Qwen-Image 本身的图像分布和风格。 请参考[我们的示例代码](./examples/qwen_image/model_inference_low_vram/Qwen-Image-EliGen-V2.py)。 diff --git a/diffsynth/pipelines/qwen_image.py b/diffsynth/pipelines/qwen_image.py index 18183e5..4e77839 100644 --- a/diffsynth/pipelines/qwen_image.py +++ b/diffsynth/pipelines/qwen_image.py @@ -150,6 +150,17 @@ class QwenImagePipeline(BasePipeline): return loss + def direct_distill_loss(self, **inputs): + self.scheduler.set_timesteps(inputs["num_inference_steps"]) + models = {name: getattr(self, name) for name in self.in_iteration_models} + for progress_id, timestep in enumerate(self.scheduler.timesteps): + timestep = timestep.unsqueeze(0).to(dtype=self.torch_dtype, device=self.device) + noise_pred = self.model_fn(**models, **inputs, timestep=timestep, progress_id=progress_id) + inputs["latents"] = self.step(self.scheduler, progress_id=progress_id, noise_pred=noise_pred, **inputs) + loss = torch.nn.functional.mse_loss(inputs["latents"].float(), inputs["input_latents"].float()) + return loss + + def _enable_fp8_lora_training(self, dtype): from transformers.models.qwen2_5_vl.modeling_qwen2_5_vl import Qwen2_5_VLRotaryEmbedding, Qwen2RMSNorm, Qwen2_5_VisionPatchEmbed, Qwen2_5_VisionRotaryEmbedding from ..models.qwen_image_dit import RMSNorm diff --git a/examples/qwen_image/README.md b/examples/qwen_image/README.md index 15e4df9..d01da52 100644 --- a/examples/qwen_image/README.md +++ b/examples/qwen_image/README.md @@ -48,7 +48,7 @@ image.save("image.jpg") |[Qwen/Qwen-Image](https://www.modelscope.cn/models/Qwen/Qwen-Image)|[code](./model_inference/Qwen-Image.py)|[code](./model_inference_low_vram/Qwen-Image.py)|[code](./model_training/full/Qwen-Image.sh)|[code](./model_training/validate_full/Qwen-Image.py)|[code](./model_training/lora/Qwen-Image.sh)|[code](./model_training/validate_lora/Qwen-Image.py)| |[Qwen/Qwen-Image-Edit](https://www.modelscope.cn/models/Qwen/Qwen-Image-Edit)|[code](./model_inference/Qwen-Image-Edit.py)|[code](./model_inference_low_vram/Qwen-Image-Edit.py)|[code](./model_training/full/Qwen-Image-Edit.sh)|[code](./model_training/validate_full/Qwen-Image-Edit.py)|[code](./model_training/lora/Qwen-Image-Edit.sh)|[code](./model_training/validate_lora/Qwen-Image-Edit.py)| |[DiffSynth-Studio/Qwen-Image-Distill-Full](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-Full)|[code](./model_inference/Qwen-Image-Distill-Full.py)|[code](./model_inference_low_vram/Qwen-Image-Distill-Full.py)|[code](./model_training/full/Qwen-Image-Distill-Full.sh)|[code](./model_training/validate_full/Qwen-Image-Distill-Full.py)|[code](./model_training/lora/Qwen-Image-Distill-Full.sh)|[code](./model_training/validate_lora/Qwen-Image-Distill-Full.py)| -|[DiffSynth-Studio/Qwen-Image-Distill-LoRA](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-LoRA)|[code](./model_inference/Qwen-Image-Distill-LoRA.py)|[code](./model_inference_low_vram/Qwen-Image-Distill-LoRA.py)|-|-|-|-| +|[DiffSynth-Studio/Qwen-Image-Distill-LoRA](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-LoRA)|[code](./model_inference/Qwen-Image-Distill-LoRA.py)|[code](./model_inference_low_vram/Qwen-Image-Distill-LoRA.py)|-|-|[code](./model_training/lora/Qwen-Image-Distill-LoRA.sh)|[code](./model_training/validate_lora/Qwen-Image-Distill-LoRA.py)| |[DiffSynth-Studio/Qwen-Image-EliGen](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen)|[code](./model_inference/Qwen-Image-EliGen.py)|[code](./model_inference_low_vram/Qwen-Image-EliGen.py)|-|-|[code](./model_training/lora/Qwen-Image-EliGen.sh)|[code](./model_training/validate_lora/Qwen-Image-EliGen.py)| |[DiffSynth-Studio/Qwen-Image-EliGen-V2](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen-V2)|[code](./model_inference/Qwen-Image-EliGen-V2.py)|[code](./model_inference_low_vram/Qwen-Image-EliGen-V2.py)|-|-|[code](./model_training/lora/Qwen-Image-EliGen.sh)|[code](./model_training/validate_lora/Qwen-Image-EliGen.py)| |[DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Canny](https://modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Canny)|[code](./model_inference/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./model_inference_low_vram/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./model_training/full/Qwen-Image-Blockwise-ControlNet-Canny.sh)|[code](./model_training/validate_full/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./model_training/lora/Qwen-Image-Blockwise-ControlNet-Canny.sh)|[code](./model_training/validate_lora/Qwen-Image-Blockwise-ControlNet-Canny.py)| diff --git a/examples/qwen_image/README_zh.md b/examples/qwen_image/README_zh.md index 97ab4c9..a7f75e4 100644 --- a/examples/qwen_image/README_zh.md +++ b/examples/qwen_image/README_zh.md @@ -48,7 +48,7 @@ image.save("image.jpg") |[Qwen/Qwen-Image](https://www.modelscope.cn/models/Qwen/Qwen-Image)|[code](./model_inference/Qwen-Image.py)|[code](./model_inference_low_vram/Qwen-Image.py)|[code](./model_training/full/Qwen-Image.sh)|[code](./model_training/validate_full/Qwen-Image.py)|[code](./model_training/lora/Qwen-Image.sh)|[code](./model_training/validate_lora/Qwen-Image.py)| |[Qwen/Qwen-Image-Edit](https://www.modelscope.cn/models/Qwen/Qwen-Image-Edit)|[code](./model_inference/Qwen-Image-Edit.py)|[code](./model_inference_low_vram/Qwen-Image-Edit.py)|[code](./model_training/full/Qwen-Image-Edit.sh)|[code](./model_training/validate_full/Qwen-Image-Edit.py)|[code](./model_training/lora/Qwen-Image-Edit.sh)|[code](./model_training/validate_lora/Qwen-Image-Edit.py)| |[DiffSynth-Studio/Qwen-Image-Distill-Full](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-Full)|[code](./model_inference/Qwen-Image-Distill-Full.py)|[code](./model_inference_low_vram/Qwen-Image-Distill-Full.py)|[code](./model_training/full/Qwen-Image-Distill-Full.sh)|[code](./model_training/validate_full/Qwen-Image-Distill-Full.py)|[code](./model_training/lora/Qwen-Image-Distill-Full.sh)|[code](./model_training/validate_lora/Qwen-Image-Distill-Full.py)| -|[DiffSynth-Studio/Qwen-Image-Distill-LoRA](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-LoRA)|[code](./model_inference/Qwen-Image-Distill-LoRA.py)|[code](./model_inference_low_vram/Qwen-Image-Distill-LoRA.py)|-|-|-|-| +|[DiffSynth-Studio/Qwen-Image-Distill-LoRA](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-LoRA)|[code](./model_inference/Qwen-Image-Distill-LoRA.py)|[code](./model_inference_low_vram/Qwen-Image-Distill-LoRA.py)|-|-|[code](./model_training/lora/Qwen-Image-Distill-LoRA.sh)|[code](./model_training/validate_lora/Qwen-Image-Distill-LoRA.py)| |[DiffSynth-Studio/Qwen-Image-EliGen](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen)|[code](./model_inference/Qwen-Image-EliGen.py)|[code](./model_inference_low_vram/Qwen-Image-EliGen.py)|-|-|[code](./model_training/lora/Qwen-Image-EliGen.sh)|[code](./model_training/validate_lora/Qwen-Image-EliGen.py)| |[DiffSynth-Studio/Qwen-Image-EliGen-V2](https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-EliGen-V2)|[code](./model_inference/Qwen-Image-EliGen-V2.py)|[code](./model_inference_low_vram/Qwen-Image-EliGen-V2.py)|-|-|[code](./model_training/lora/Qwen-Image-EliGen.sh)|[code](./model_training/validate_lora/Qwen-Image-EliGen.py)| |[DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Canny](https://modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Blockwise-ControlNet-Canny)|[code](./model_inference/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./model_inference_low_vram/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./model_training/full/Qwen-Image-Blockwise-ControlNet-Canny.sh)|[code](./model_training/validate_full/Qwen-Image-Blockwise-ControlNet-Canny.py)|[code](./model_training/lora/Qwen-Image-Blockwise-ControlNet-Canny.sh)|[code](./model_training/validate_lora/Qwen-Image-Blockwise-ControlNet-Canny.py)| diff --git a/examples/qwen_image/model_training/lora/Qwen-Image-Distill-LoRA.sh b/examples/qwen_image/model_training/lora/Qwen-Image-Distill-LoRA.sh new file mode 100644 index 0000000..061bebb --- /dev/null +++ b/examples/qwen_image/model_training/lora/Qwen-Image-Distill-LoRA.sh @@ -0,0 +1,24 @@ +accelerate launch examples/qwen_image/model_training/train.py \ + --dataset_base_path data/example_image_dataset \ + --dataset_metadata_path data/example_image_dataset/metadata_distill_qwen_image.csv \ + --data_file_keys "image" \ + --extra_inputs "seed,rand_device,num_inference_steps,cfg_scale" \ + --height 1328 \ + --width 1328 \ + --dataset_repeat 50 \ + --model_id_with_origin_paths "Qwen/Qwen-Image:transformer/diffusion_pytorch_model*.safetensors,Qwen/Qwen-Image:text_encoder/model*.safetensors,Qwen/Qwen-Image:vae/diffusion_pytorch_model.safetensors" \ + --learning_rate 1e-4 \ + --num_epochs 5 \ + --remove_prefix_in_ckpt "pipe.dit." \ + --output_path "./models/train/Qwen-Image-Distill-LoRA_lora" \ + --lora_base_model "dit" \ + --lora_target_modules "to_q,to_k,to_v,add_q_proj,add_k_proj,add_v_proj,to_out.0,to_add_out,img_mlp.net.2,img_mod.1,txt_mlp.net.2,txt_mod.1" \ + --lora_rank 32 \ + --use_gradient_checkpointing \ + --dataset_num_workers 8 \ + --find_unused_parameters \ + --task direct_distill + +# This is an experimental training feature designed to directly distill the model, enabling generation results with fewer steps to approximate those achieved with more steps. +# The model (https://www.modelscope.cn/models/DiffSynth-Studio/Qwen-Image-Distill-LoRA) is trained using this script. +# The sample dataset is provided solely to demonstrate the dataset format. For actual usage, please construct a larger dataset using the base model. diff --git a/examples/qwen_image/model_training/train.py b/examples/qwen_image/model_training/train.py index 1370b32..1b4faec 100644 --- a/examples/qwen_image/model_training/train.py +++ b/examples/qwen_image/model_training/train.py @@ -19,6 +19,7 @@ class QwenImageTrainingModule(DiffusionTrainingModule): use_gradient_checkpointing_offload=False, extra_inputs=None, enable_fp8_training=False, + task="sft", ): super().__init__() # Load models @@ -38,6 +39,7 @@ class QwenImageTrainingModule(DiffusionTrainingModule): 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.task = task def forward_preprocess(self, data): @@ -82,11 +84,21 @@ class QwenImageTrainingModule(DiffusionTrainingModule): def forward(self, data, inputs=None, return_inputs=False): + # Inputs 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) + + # Loss + if self.task == "sft": + models = {name: getattr(self.pipe, name) for name in self.pipe.in_iteration_models} + loss = self.pipe.training_loss(**models, **inputs) + elif self.task == "data_process": + loss = inputs + elif self.task == "direct_distill": + loss = self.pipe.direct_distill_loss(**inputs) + else: + raise NotImplementedError(f"Unsupported task: {self.task}.") return loss @@ -122,10 +134,12 @@ if __name__ == "__main__": use_gradient_checkpointing_offload=args.use_gradient_checkpointing_offload, extra_inputs=args.extra_inputs, enable_fp8_training=args.enable_fp8_training, + task=args.task, ) model_logger = ModelLogger(args.output_path, remove_prefix_in_ckpt=args.remove_prefix_in_ckpt) launcher_map = { "sft": launch_training_task, - "data_process": launch_data_process_task + "data_process": launch_data_process_task, + "direct_distill": launch_training_task, } launcher_map[args.task](dataset, model, model_logger, args=args) diff --git a/examples/qwen_image/model_training/validate_lora/Qwen-Image-Distill-LoRA.py b/examples/qwen_image/model_training/validate_lora/Qwen-Image-Distill-LoRA.py new file mode 100644 index 0000000..d56b9af --- /dev/null +++ b/examples/qwen_image/model_training/validate_lora/Qwen-Image-Distill-LoRA.py @@ -0,0 +1,23 @@ +from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig +import torch + + +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"), + ], + tokenizer_config=ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="tokenizer/"), +) +pipe.load_lora(pipe.dit, "models/train/Qwen-Image-Distill-LoRA_lora/epoch-4.safetensors") +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 = pipe( + prompt, + seed=0, + num_inference_steps=4, + cfg_scale=1, +) +image.save("image.jpg")