Missing diffusion_pytorch_model.bin File in Pretrained Weights Directory

[unsiao]

Description:

I encountered an error when trying to load a pretrained model using the UNet2DConditionModel.from_pretrained method. The error message indicates that the diffusion_pytorch_model.bin file is not found in the specified directory.

Error Message:

OSError: Error no file named diffusion_pytorch_model.bin found in directory ./pretrained_weights/sd-image-variations-diffusers.

Directory Structure:

(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights/sd-image-variations-diffusers# ls
README.md            earring.jpg        inputs.jpg        scheduler       v2-montage.jpg
alias-montage.jpg    feature_extractor  model_index.json  unet            vae
default-montage.jpg  image_encoder      safety_checker    v1-montage.jpg
(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights/sd-image-variations-diffusers# cd ../
(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights# ls
README.md        configuration.json  motion_module.pth  reference_unet.pth             sd-vae-ft-mse
audio_processor  denoising_unet.pth  pose_encoder.pth   sd-image-variations-diffusers
(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights# ls -l
total 5805204
-rw-r--r-- 1 root root      12049 Dec  5 12:10 README.md
drwxr-xr-x 2 root root       4096 Dec  5 12:24 audio_processor
-rw-r--r-- 1 root root         50 Dec  5 12:10 configuration.json
-rw-r--r-- 1 root root 1702192597 Dec  5 12:16 denoising_unet.pth
-rw-r--r-- 1 root root  908951917 Dec  5 12:13 motion_module.pth
-rw-r--r-- 1 root root 1702048079 Dec  5 12:18 pose_encoder.pth
-rw-r--r-- 1 root root 1631280213 Dec  5 12:16 reference_unet.pth
drwxr-xr-x 9 root root       4096 Dec  5 12:22 sd-image-variations-diffusers
drwxr-xr-x 3 root root       4096 Dec  5 12:21 sd-vae-ft-mse
(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights# ls sd-image-variations-diffusers/vae/
config.json  diffusion_pytorch_model.bin

Steps to Reproduce:

1. Run the script app.py.
2. Observe the error message indicating the missing diffusion_pytorch_model.bin file.

Expected Behavior:
The script should successfully load the pretrained model without any errors.

Actual Behavior:
The script fails to load the pretrained model and raises an OSError indicating that the diffusion_pytorch_model.bin file is not found.

Possible Solution:
Update the path in the code to point to the correct directory where the diffusion_pytorch_model.bin file is located. For example:

reference_unet = UNet2DConditionModel.from_pretrained("./pretrained_weights/sd-image-variations-diffusers/vae", subfolder="unet", use_safetensors=False).to(dtype=dtype, device=device)

Additional Information:

• Python Version: 3.10
• Operating System: Ubuntu20.04

描述:

在尝试使用 UNet2DConditionModel.from_pretrained 方法加载预训练模型时，遇到了错误。错误信息表明在指定的目录中找不到 diffusion_pytorch_model.bin 文件。

错误信息:

OSError: Error no file named diffusion_pytorch_model.bin found in directory ./pretrained_weights/sd-image-variations-diffusers.

目录结构:

(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights/sd-image-variations-diffusers# ls
README.md            earring.jpg        inputs.jpg        scheduler       v2-montage.jpg
alias-montage.jpg    feature_extractor  model_index.json  unet            vae
default-montage.jpg  image_encoder      safety_checker    v1-montage.jpg
(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights/sd-image-variations-diffusers# cd ../
(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights# ls
README.md        configuration.json  motion_module.pth  reference_unet.pth             sd-vae-ft-mse
audio_processor  denoising_unet.pth  pose_encoder.pth   sd-image-variations-diffusers
(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights# ls -l
total 5805204
-rw-r--r-- 1 root root      12049 Dec  5 12:10 README.md
drwxr-xr-x 2 root root       4096 Dec  5 12:24 audio_processor
-rw-r--r-- 1 root root         50 Dec  5 12:10 configuration.json
-rw-r--r-- 1 root root 1702192597 Dec  5 12:16 denoising_unet.pth
-rw-r--r-- 1 root root  908951917 Dec  5 12:13 motion_module.pth
-rw-r--r-- 1 root root 1702048079 Dec  5 12:18 pose_encoder.pth
-rw-r--r-- 1 root root 1631280213 Dec  5 12:16 reference_unet.pth
drwxr-xr-x 9 root root       4096 Dec  5 12:22 sd-image-variations-diffusers
drwxr-xr-x 3 root root       4096 Dec  5 12:21 sd-vae-ft-mse
(base) root@VM-3-148-ubuntu:/workspace/echomimic_v2/pretrained_weights# ls sd-image-variations-diffusers/vae/
config.json  diffusion_pytorch_model.bin

重现步骤:

1. 运行脚本 app.py。
2. 观察错误信息，指出缺少 diffusion_pytorch_model.bin 文件。

预期行为:
脚本应成功加载预训练模型，不出现任何错误。

实际行为:
脚本无法加载预训练模型，并引发 OSError，指出找不到 diffusion_pytorch_model.bin 文件。

可能的解决方案:
更新代码中的路径，指向包含 diffusion_pytorch_model.bin 文件的正确目录。例如：

reference_unet = UNet2DConditionModel.from_pretrained("./pretrained_weights/sd-image-variations-diffusers/vae", subfolder="unet", use_safetensors=False).to(dtype=dtype, device=device)

附加信息:

• Python 版本: 3.10
• 操作系统: Ubuntu20.04LTS