Error when generating a preview image

[Trauter1]

Traceback (most recent call last):
  File "/notebooks/stable-diffusion-webui/modules/ui.py", line 185, in f
    res = list(func(*args, **kwargs))
  File "/notebooks/stable-diffusion-webui/webui.py", line 54, in f
    res = func(*args, **kwargs)
  File "/notebooks/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/dreambooth.py", line 256, in start_training
    trained_steps = main(config)
  File "/notebooks/stable-diffusion-webui/extensions/sd_dreambooth_extension/dreambooth/train_dreambooth.py", line 829, in main
    g_cuda = torch.Generator(device=accelerator.device).manual_seed(args.seed)
AttributeError: 'DreamboothConfig' object has no attribute 'seed'

Worked fine before updating to the latest version.

[gonjay]

Change this line

parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")

to:

parser.add_argument("--seed", type=int, default=2333, help="A seed for reproducible training.")

Then reload webui

[ThereforeGames]

I'm receiving the same error after updating train_dreambooth.py as @gonjay suggested. Seems like another edit is needed somewhere?

[Trauter1]

Sadly, it doesn't work for me. Im running paperspace notebook it that matters. @gonjay

[rabidcopy]

Same.

Traceback (most recent call last):
  File "/notebooks/SDW/modules/ui.py", line 180, in f
    res = list(func(*args, **kwargs))
  File "/notebooks/SDW/webui.py", line 55, in f
    res = func(*args, **kwargs)
  File "/notebooks/SDW/extensions/sd_dreambooth_extension/dreambooth/dreambooth.py", line 256, in start_training
    trained_steps = main(config)
  File "/notebooks/SDW/extensions/sd_dreambooth_extension/dreambooth/train_dreambooth.py", line 829, in main
    g_cuda = torch.Generator(device=accelerator.device).manual_seed(args.seed)

Edit: Upon trying to manually hardcode a seed, I then run into this.

Traceback (most recent call last):
  File "/notebooks/SDW/modules/ui.py", line 180, in f
    res = list(func(*args, **kwargs))
  File "/notebooks/SDW/webui.py", line 55, in f
    res = func(*args, **kwargs)
  File "/notebooks/SDW/extensions/sd_dreambooth_extension/dreambooth/dreambooth.py", line 256, in start_training
    trained_steps = main(config)
  File "/notebooks/SDW/extensions/sd_dreambooth_extension/dreambooth/train_dreambooth.py", line 831, in main~
    image = pipeline(
  File "/usr/local/lib/python3.9/dist-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "/usr/local/lib/python3.9/dist-packages/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", line 395, in __call__
    latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
  File "/usr/local/lib/python3.9/dist-packages/diffusers/schedulers/scheduling_pndm.py", line 220, in step
    return self.step_plms(model_output=model_output, timestep=timestep, sample=sample, return_dict=return_dict)
  File "/usr/local/lib/python3.9/dist-packages/diffusers/schedulers/scheduling_pndm.py", line 336, in step_plms
    prev_sample = self._get_prev_sample(sample, timestep, prev_timestep, model_output)
  File "/usr/local/lib/python3.9/dist-packages/diffusers/schedulers/scheduling_pndm.py", line 371, in _get_prev_sample
    alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
IndexError: tensors used as indices must be long, byte or bool tensors

Reverting to a354abd is the only way I can train with preview images, but then they are just the base instance prompts again and generation settings aren't used. I tried to do the old pipeline with the args for generation parameters and prompt input but still a no-go.

I small workaround for the instance prompt issue is to change line 829 of train_dreambooth.py in this older commit to prompt = f'{args.instance_prompt} {args.class_prompt}' so that at least generates previews with the full instance+class prompt so it's a better look at how your training is going.

[Dawgmastah]

Can confirm the issue, last night preview images worked fine

[entmike]

Same issue I just reported here (didn't notice you already had one open here based on subject line) #30

[rabidcopy]

Possibly fixed by 43ec3b6. Can't test right now.
Edit: Still getting the same error, I think there's still a problem with there being no default seed specified in the argument parser?

Traceback (most recent call last):
  File "/notebooks/SDW/modules/ui.py", line 180, in f
    res = list(func(*args, **kwargs))
  File "/notebooks/SDW/webui.py", line 55, in f
    res = func(*args, **kwargs)
  File "/notebooks/SDW/extensions/sd_dreambooth_extension/dreambooth/dreambooth.py", line 256, in start_training
    trained_steps = main(config)
  File "/notebooks/SDW/extensions/sd_dreambooth_extension/dreambooth/train_dreambooth.py", line 451, in main
    if args.seed is not None:
AttributeError: 'DreamboothConfig' object has no attribute 'seed'

Changing the --seed line to parser.add_argument("--seed", type=int, default=5013, help="A seed for reproducible training.") still produces the same error claiming there isn't an attribute.

Manually setting a seed with args.seed = int(5013) just before line 451 gets past that error but again throws this same error from before.

Traceback (most recent call last):
  File "/notebooks/SDW/modules/ui.py", line 180, in f
    res = list(func(*args, **kwargs))
  File "/notebooks/SDW/webui.py", line 55, in f
    res = func(*args, **kwargs)
  File "/notebooks/SDW/extensions/sd_dreambooth_extension/dreambooth/dreambooth.py", line 256, in start_training
    trained_steps = main(config)
  File "/notebooks/SDW/extensions/sd_dreambooth_extension/dreambooth/train_dreambooth.py", line 838, in main
    image = pipeline(
  File "/usr/local/lib/python3.9/dist-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
    return func(*args, **kwargs)
  File "/usr/local/lib/python3.9/dist-packages/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py", line 334, in __call__
    latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
  File "/usr/local/lib/python3.9/dist-packages/diffusers/schedulers/scheduling_pndm.py", line 217, in step
    return self.step_plms(model_output=model_output, timestep=timestep, sample=sample, return_dict=return_dict)
  File "/usr/local/lib/python3.9/dist-packages/diffusers/schedulers/scheduling_pndm.py", line 332, in step_plms
    prev_sample = self._get_prev_sample(sample, timestep, prev_timestep, model_output)
  File "/usr/local/lib/python3.9/dist-packages/diffusers/schedulers/scheduling_pndm.py", line 367, in _get_prev_sample
    alpha_prod_t_prev = self.alphas_cumprod[prev_timestep] if prev_timestep >= 0 else self.final_alpha_cumprod
IndexError: tensors used as indices must be long, byte or bool tensors

I don't know if anyone else is experiencing this problem. I guess for now I'm going to stick to an older commit (a354abd) and change the preview generation section at around 820 lines into train_dreambooth.py to this.

                            if save_img:
                                shared.state.textinfo = "Generating preview..."
                                pipeline.safety_checker = dumb_safety
                                prompt = f'{args.instance_prompt} {args.save_sample_prompt}'
                                negativeprompt = f'{args.save_sample_negative_prompt}'
                                last_saved_image = os.path.join(args.output_dir, "logging",
                                                                f'{prompt}_{lifetime_step}.png')
                                image = pipeline(prompt, negative_prompt=negativeprompt, num_inference_steps=40, guidance_scale=7.5).images[0]
                                shared.state.current_image = image
                                image.save(last_saved_image)

Seems to work and passes positive and negative prompts on to the generation. Didn't really bother with plugging in steps and scale though.

Edit: I think the IndexError: tensors used as indices must be long, byte or bool tensors is being caused by trying to use a negative prompt for preview generation.

[d8ahazard]

Think I got it now:

9e6ebef

[Wakzor]

Im still having an error when generating preview. CPU: False Adam: True, Prec: bf16, Prior: False, Grad: True, TextTr: True
File "C:\Users\user\stable-diffusion-webui\modules\ui.py", line 185, in f res = list(func(*args, **kwargs)) File "C:\Users\user\stable-diffusion-webui\webui.py", line 54, in f res = func(*args, **kwargs) File "C:\Users\user\stable-diffusion-webui\extensions\sd_dreambooth_extension\dreambooth\dreambooth.py", line 265, in start_training trained_steps = main(config) File "C:\Users\user\stable-diffusion-webui\extensions\sd_dreambooth_extension\dreambooth\train_dreambooth.py", line 829, in main revision=args.revision, AttributeError: 'DreamboothConfig' object has no attribute 'revision'

[f8upd8]

Still broken as of right now (IndexError: tensors used as indices must be long, byte or bool tensors)

[d8ahazard]

This issue has been resolved.