Error on MPS: AttnProcessor2_0.__call__() got an unexpected keyword argument 'real_attn_probs'

[jwooldridge234]

Hey, just tried running and I got this error when running runwayml/stable-diffusion-v1-5. Happened after restarting as well. I'm running the latest Auto1111 version on MacOS. Let me know if there's anything else I can do to help debug.

Traceback (most recent call last):
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/routes.py", line 488, in run_predict
    output = await app.get_blocks().process_api(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1431, in process_api
    result = await self.call_function(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1103, in call_function
    prediction = await anyio.to_thread.run_sync(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/anyio/to_thread.py", line 33, in run_sync
    return await get_asynclib().run_sync_in_worker_thread(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/anyio/_backends/_asyncio.py", line 877, in run_sync_in_worker_thread
    return await future
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/anyio/_backends/_asyncio.py", line 807, in run
    result = context.run(func, *args)
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/utils.py", line 707, in wrapper
    response = f(*args, **kwargs)
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/rich_text_on_tab.py", line 146, in generate
    plain_img = model.produce_attn_maps([base_text_prompt], [negative_text],
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/models/region_diffusion.py", line 214, in produce_attn_maps
    noise_pred = self.unet(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/models/unet_2d_condition.py", line 908, in forward
    sample, res_samples = downsample_block(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/models/unet_2d_blocks.py", line 999, in forward
    hidden_states = attn(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/models/transformer_2d.py", line 291, in forward
    hidden_states = block(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/models/attention.py", line 155, in forward
    attn_output, _ = self.attn1(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1538, in _call_impl
    result = forward_call(*args, **kwargs)
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/models/attention_processor.py", line 330, in forward
    return self.processor(
TypeError: AttnProcessor2_0.__call__() got an unexpected keyword argument 'real_attn_probs'

[jwooldridge234]

Not sure if it's related, but I had to manually install "scikit-learn" rather than "sklearn" as the latter isn't accessible on MPS (deprecated).

[songweige]

You are doing the right thing with scikit-learn! I think the error is related to this issue (https://github.com/songweige/rich-text-to-image/issues/11). Could you check the version of diffusers package you are using?

[songweige]

I think I might know the error. Are you using pytorch with version >= 2.0?

[jwooldridge234]

I think I might know the error. Are you using pytorch with version >= 2.0?

Yeah. Do I need to down/upgrade?

[jwooldridge234]

I'm running Diffusers 0.18.2.

[songweige]

Let me see if I can reproduce and fix it on my end first!

[jwooldridge234]

Sounds great, thanks!

[songweige]

Hey @jwooldridge234, I was able to reproduce the error and make a fix for using pytoch>=2.0. It should work now. Could you try to update the extension and see if the error is gone or not? Thanks!

[jwooldridge234]

@songweige It is! Thanks for getting that working.

Unfortunately hit another error:

text_input {"ops":[{"insert":"A stunning photo of a "},{"attributes":{"link":"http://localhost:7860/file=./extensions/sd-webui-rich-text/cat%20wearing%20a%20red%20hat%20and%20a%20green%20sweater"},"insert":"cat"},{"insert":" and a "},{"attributes":{"link":"http://localhost:7860/file=./extensions/sd-webui-rich-text/blue%20dog%20with%20yellow%20eyes"},"insert":"dog"},{"insert":".\n"}]} 512 512 41 8.5 9 0.45 0 0.3 0.5 
/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/models/region_diffusion.py:202: FutureWarning: Accessing config attribute `in_channels` directly via 'UNet2DConditionModel' object attribute is deprecated. Please access 'in_channels' over 'UNet2DConditionModel's config object instead, e.g. 'unet.config.in_channels'.
  (text_embeddings.shape[0] // 2, self.unet.in_channels, height // 8, width // 8), device=self.device)
time lapses to get attention maps: 341.7217
Traceback (most recent call last):
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/routes.py", line 488, in run_predict
    output = await app.get_blocks().process_api(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1431, in process_api
    result = await self.call_function(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/blocks.py", line 1103, in call_function
    prediction = await anyio.to_thread.run_sync(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/anyio/to_thread.py", line 33, in run_sync
    return await get_asynclib().run_sync_in_worker_thread(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/anyio/_backends/_asyncio.py", line 877, in run_sync_in_worker_thread
    return await future
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/anyio/_backends/_asyncio.py", line 807, in run
    result = context.run(func, *args)
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/gradio/utils.py", line 707, in wrapper
    response = f(*args, **kwargs)
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/rich_text_on_tab.py", line 156, in generate
    color_obj_masks, segments_vis, token_maps = get_token_maps(model.selfattn_maps, model.crossattn_maps, model.n_maps, run_dir,
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/models/utils/attention_utils.py", line 722, in get_token_maps
    resized_token_maps = [token_map.unsqueeze(1).repeat(
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/extensions/sd-webui-rich-text/scripts/models/utils/attention_utils.py", line 723, in <listcomp>
    [1, 4, 1, 1]).to(attn_map.dtype).cuda() for token_map in resized_token_maps]
  File "/Users/jackwooldridge/StableDiffusion/stable-diffusion-webui/venv/lib/python3.10/site-packages/torch/cuda/__init__.py", line 239, in _lazy_init
    raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

What are the hardware requirements for this method btw? It took 6 minutes to get the attention maps. I have 8GB vram.

[jwooldridge234]

@songweige Got past that by casting to "mps" whenever it specified "cuda" and changing the .cuda() calls to .to('mps'). The main problem is how slow it is at this point. It gets past the attention map stage after ~6-7 minutes and then moves to (I assume) the token maps. I finally killed it after it ran for over 20 minutes with no change. Maybe my GPU just isn't powerful enough to handle it?

[songweige]

Glad it works!!! And thanks for debugging the code. I didn't consider the scenario of running it on mac before.

Regarding the time, it sounds like you have done the plain-text-to-image part and moved to the rich-text-to-image part. The time of that part depends on what and how many rich-text formats you use in the text prompt. Usually, font, footnote, and font size are the fastest. Color is a lot slower as it requires some gradient computation. Maybe try to reduce the steps? Also, I'm curious how long does it take for you to run normal Stable Diffusion sampling in your setting?

[jwooldridge234]

@songweige I will try running without any rich text and reducing the steps, and see if I can get the gen time down. Thanks!

[jwooldridge234]

Going to close this as I believe you resolved all the issues I was having. :)