stable_diffusion模型报错

[metaire]

1）PaddleHub2.1.0，PaddlePaddle2.1.2.post112 2）系统环境：Windows。python版本：3.7

• 复现信息：如为报错，请给出复现环境、复现步骤
• 运行出错，运行方式：
• module = hub.Module(name="stable_diffusion_img2img") text_prompts = ["A fantasy landscape, trending on artstation"] da = module.generate_image(text_prompts=text_prompts, init_image='./image/img.png', batch_size=2, output_dir='./stable_diffusion_img2img_out/') da[0].chunks[-1].chunks.plot_image_sprites(skip_empty=True, show_index=True, keep_aspect_ratio=True) da[0].chunks[-1].chunks.save_gif('stable_diffusion_img2img_out-merged-result.gif') da[0].chunks[0].chunks.plot_image_sprites(skip_empty=True, show_index=True, keep_aspect_ratio=True) da[0].chunks[0].chunks.save_gif('stable_diffusion_img2img-image-0-result.gif') 错误： Traceback (most recent call last): File "D:/metaire/stable_diffusion/imgtoimg.py", line 12, in output_dir='./stable_diffusion_img2img_out/') File "D:\Anaconda3\lib\site-packages\paddlehub\compat\paddle_utils.py", line 220, in runner return func(*args, **kwargs) File "C:\Users\Administrator.paddlehub\modules\stable_diffusion_img2img\module.py", line 205, in generate_image init_latents = self.scheduler.add_noise(init_latents, noise, timesteps) File "C:\Users\Administrator.paddlehub\modules\stable_diffusion_img2img\diffusers\schedulers\scheduling_pndm.py", line 255, in add_noise sqrt_alpha_prod = self.match_shape(sqrt_alpha_prod, original_samples) File "C:\Users\Administrator.paddlehub\modules\stable_diffusion_img2img\diffusers\schedulers\scheduling_utils.py", line 73, in match_shape values = values[..., None] File "D:\Anaconda3\lib\site-packages\paddle\fluid\dygraph\varbase_patch_methods.py", line 569, in getitem return self._getitem_index_not_tensor(item) ValueError: (InvalidArgument) too many indices (2) for tensor of dimension 1 [Hint: Expected size <= rank == true, but received size <= rank:0 != true:1.] (at C:\home\workspace\Paddle_release2\paddle\fluid\pybind\imperative.cc:421)

[jm12138]

可以尝试升级至最新版本的 PaddlePaddle 和 PaddleHub，还有检查一下图片文件，我这边测试是能够正常运行的

[VVsxmja]

可以把 modules/image/text_to_image/stable_diffusion/README.md 中的环境依赖改为 paddlepaddle>=2.2 吗？

我用 2.1 版本会报一样的错，升级到 2.2 及以上就可以用了。

问题应该出在 modules/image/text_to_image/stable_diffusion/module.py 第179行

paddle.amp.auto_cast(enable=enable_fp16, level='O2')

这里 level 这个参数是 paddlepaddle 2.2 之后才有的

[hjj-lmx]

stable_diffusion_img2img需要多大的显存，我现在24g的都会报显存不足