v0xie / sd-webui-agentattention Goto Github PK

I generated an image using Apply to Hires. Fix Only option and at first the speed was fine, but after the first pass the generation became very slow, from around 9 seconds to a minute and 20 seconds. Tried unloading and reloading the model, changing the model, but it still remained at that speed.

7:12:36-721823 ERROR Hypertile error: width=512 height=768 xformers_attention_forward() got an unexpected keyword
argument 'agent'
07:12:36-726157 ERROR Exception: xformers_attention_forward() got an unexpected keyword argument 'agent'
...
07:12:36-756215 ERROR gradio call: TypeError
╭────────────────────────────── Traceback (most recent call last) ─────────────────────────────╮
│ C:\AI\SDNEXT\automatic\modules\sd_hijack_hypertile.py:112 in wrapper
││
││ 111 │ │ │ │ │ │ x = rearrange(x, "b (nh h nw w) c -> (b nh nw) (h w) c", h=h //
││ ❱ 112 │ │ │ │ │ out = forward(x, *args[1:], **kwargs)
││ 113 │ │ │ │ │ if do_split: │╰───────────────────────────────────────────────────────────────────────────────╯
TypeError: xformers_attention_forward() got an unexpected keyword argument 'agent'

During handling of the above exception, another exception occurred:

╭───────────────────────────── Traceback (most recent call last) ──────────────────────────────╮
│ C:\AI\SDNEXT\automatic\modules\call_queue.py:31 in f
││
││ 30 │ │ │ try:
││ ❱ 31 │ │ │ │ res = func(*args, **kwargs)
││ 32 │ │ │ │ progress.record_results(id_task, res)
││
││ C:\AI\SDNEXT\automatic\modules\txt2img.py:88 in txt2img
││
││ 87 │ if processed is None:
││ ❱ 88 │ │ processed = processing.process_images(p)
││ 89 │ p.close()
││
││ C:\AI\SDNEXT\automatic\modules\processing.py:760 in process_images
││
││ 759 │ │ │ with context_hypertile_vae(p), context_hypertile_unet(p):
││ ❱ 760 │ │ │ │ res = process_images_inner(p)
││ 761
││
││ C:\AI\SDNEXT\automatic\extensions-builtin\sd-webui-controlnet\scripts\batch_hijack.py:42 in
││ processing_process_images_hijack
││
││ 41 │ │ │ # we are not in batch mode, fallback to original function
││ ❱ 42 │ │ │ return getattr(processing, '__controlnet_original_process_images_inner')(p,
││ 43
││
││ C:\AI\SDNEXT\automatic\modules\processing.py:900 in process_images_inner
││
││ 899 │ │ │ │ with devices.without_autocast() if devices.unet_needs_upcast else device
││ ❱ 900 │ │ │ │ │ samples_ddim = p.sample(conditioning=c, unconditional_conditioning=u
││ 901 │ │ │ │ x_samples_ddim = [decode_first_stage(p.sd_model, samples_ddim[i:i+1].to(
││
││ ... 36 frames hidden ...
││
││ C:\AI\SDNEXT\automatic\venv\lib\site-packages\torch\utils\checkpoint.py:230 in forward
││
││ 229 │ │ with torch.no_grad():
││ ❱ 230 │ │ │ outputs = run_function(*args)
││ 231 │ │ return outputs
││
││ C:\AI\SDNEXT\automatic\extensions\sd-webui-agentattention-SDNEXT\agentsd\patch.py:78 in _forward
││
││ 77 │ │ │ feature, agent = m_a(y)
││ ❱ 78 │ │ │ x = u_a(self.attn1(feature, agent=agent, context=context if self.disable_sel
││ 79 │ │ │ y = self.norm2(x)
││
││ C:\AI\SDNEXT\automatic\venv\lib\site-packages\torch\nn\modules\module.py:1518 in _wrapped_call_impl
││
││ 1517 │ │ else:
││ ❱ 1518 │ │ │ return self._call_impl(*args, **kwargs)
││ 1519
││
││ C:\AI\SDNEXT\automatic\venv\lib\site-packages\torch\nn\modules\module.py:1527 in _call_impl
││
││ 1526 │ │ │ │ or _global_forward_hooks or _global_forward_pre_hooks):
││ ❱ 1527 │ │ │ return forward_call(*args, **kwargs)
││ 1528
││
││ C:\AI\SDNEXT\automatic\modules\sd_hijack_hypertile.py:120 in wrapper
││
││ 119 │ │ │ │ │ │ log.error(f'Hypertile error: width={width} height={height} {e}')
││ ❱ 120 │ │ │ │ │ out = forward(x, *args[1:], **kwargs)
││ 121 │ │ │ return out │╰──────────────────────────────────────────────────────────────────────────────╯
TypeError: xformers_attention_forward() got an unexpected keyword argument 'agent'

1. When Agent Attention provide accelerated image generation, what are the major tradeoffs?
2. How is this different from the goals of LCM (Latent Consistency Model)? https://github.com/0xbitches/sd-webui-lcm
3. What is with the "Swin" in the paper that also appeared in of the Waifu Diffusion Tagger (right next to ViT and ConvNext)?

I don't see any speed improvements on SD 1.5. There are improvements in quality, but not in speed.