Exporting a second model caused a UNET selection error

[banshek7]

Skip all this I was confused about the origin of my problem Lol I couldnt figure out how my new glorious 13it/s with my potato 2060 card was suddenly ruined, even when selecting my new SD_VAE option to none, but for some reason I had to actually disable it in the user interface options completely to get back to ripping my pictures quickly again.

Is there a way I can get sd vae to work with your tensor export stuff. Like could I just put it in my stable diffusion model folder and export it and then put it in VAE lol... Ugh sorry I just really want to finish this all off with that =(

[banshek7]

Im not really sure whats going on anymore, the next generation press was 4 lol. Not really sure what happened, I had tried to download a model update with civitihelper and that failed, and then i played with deforum once, i did use the selected tensor Unet model for deforum, idk how that could have scrambled it up, it made me a cool video pretty quickly =( Meh.

I'll have to keep troubleshooting. I'm still curious about if I have to export the VAE

[banshek7]

maybe start reading here

If I had to guess now that i've dittled with it more, the speeds going way down and being random and screwed up started when I exported realism V6

At first it seemed fine, but for instance I just set unet to automatic instead of matching my dream to the tensor dream, and then it was over 10 it/s again, I hit generate and then it was 4.

That's when to me it showed something was retarded with it picking the right unet now, so even though ive restarted a bunch, even my whole computer, i hit refresh on the unet list, manually selected dream again, and now im hitting 13it/s

super triggering but im a little comforted knowing there is probably a rhyme and a reason to whats happening and a fix.

I just comfirmed it through a bunch of testing, its brokie, first off, i realized when its slow that tensor field that prints out in the CMD isnt even coming up, even though the right unet is selected (since adding the second export model; DEV version of your program) I have to hit refresh on the UNET field, then go to automatic each generations. Really sad but i hope to figure out how to fix it lmao.

[banshek7]

Pretty good news for me, if I keep SD Unet on automatic, and hit that little refresh icon after each generation the glitch stops and I have tensorRT working again 13it/s with my asus strix poop card, adding a second model really screwed that up

[banshek7]

I'll leave you with my last generation, courtesy of your awesome work here, I hope you can fix that or warn people if its repeatable because that was mind boggling to go from 13 to 1-3 with no explanation. using SD_VAE is fine, adding that back in had no ill effect but it looks like I'll be hitting refresh on SD_UNET - Automatic forever now, not so bad considering lol

[banshek7]

I just wanted to update that it seems worse than ever now that I exported a third one, I have to do some really gypsy stuff, switching back and forth on the model and unet drop downs, hitting refresh, and then all the sudden its working again like lol. Even sometimes now its printing the tensor field thing and going slow, or I was just having this insane print in the cmd, I literally hit refresh on SD UNET one more time and that was the tipping point of it going back to working again.

To create a public link, setshare=Trueinlaunch()`. Startup time: 46.9s (prepare environment: 20.4s, import torch: 3.6s, import gradio: 1.1s, setup paths: 1.3s, initialize shared: 0.2s, other imports: 0.9s, setup codeformer: 0.2s, load scripts: 4.9s, create ui: 12.9s, gradio launch: 0.6s, app_started_callback: 0.8s). Traceback (most recent call last): File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\venv\lib\site-packages\gradio\routes.py", line 488, in run_predict output = await app.get_blocks().process_api( File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 1431, in process_api result = await self.call_function( File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\venv\lib\site-packages\gradio\blocks.py", line 1103, in call_function prediction = await anyio.to_thread.run_sync( File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\venv\lib\site-packages\anyio\to_thread.py", line 33, in run_sync return await get_asynclib().run_sync_in_worker_thread( File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\venv\lib\site-packages\anyio_backends_asyncio.py", line 877, in run_sync_in_worker_thread return await future File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\venv\lib\site-packages\anyio_backends_asyncio.py", line 807, in run result = context.run(func, args) File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\venv\lib\site-packages\gradio\utils.py", line 707, in wrapper response = f(args, **kwargs) File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\modules\ui_extra_networks.py", line 392, in pages_html return refresh() File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\modules\ui_extra_networks.py", line 400, in refresh ui.pages_contents = [pg.create_html(ui.tabname) for pg in ui.stored_extra_pages] File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\modules\ui_extra_networks.py", line 400, in ui.pages_contents = [pg.create_html(ui.tabname) for pg in ui.stored_extra_pages] File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\modules\ui_extra_networks.py", line 162, in create_html self.items = {x["name"]: x for x in self.list_items()} File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\modules\ui_extra_networks.py", line 162, in self.items = {x["name"]: x for x in self.list_items()} File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\modules\ui_extra_networks_checkpoints.py", line 35, in list_items yield self.create_item(name, index) File "C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\modules\ui_extra_networks_checkpoints.py", line 18, in create_item path, ext = os.path.splitext(checkpoint.filename) AttributeError: 'NoneType' object has no attribute 'filename' Activating unet: [TRT] ClothingAdjuster3 (realisticVisionV60B1_v60B1VAE) Loading TensorRT engine: C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\models\Unet-trt\realisticVisionV60B1_v60B1VAE_d6aa28fb_cc75_sample=1x4x64x64+2x4x64x64+8x4x128x128-timesteps=1+2+8-encoder_hidden_states=1x77x768+2x77x768+8x154x768.trt

Loaded Profile: 0 sample = [(1, 4, 64, 64), (2, 4, 64, 64), (8, 4, 128, 128)] timesteps = [(1,), (2,), (8,)] encoder_hidden_states = [(1, 77, 768), (2, 77, 768), (8, 154, 768)] latent = [(-1945890048), (-1945887232), (-1945891072)]

38%|███████████████████████████████▏ | 19/50 [00:07<00:12, 2.57it/s] Total progress: 19%|████████████▎ | 19/100 [00:19<01:23, 1.03s/it] Reusing loaded model realisticVisionV60B1_v60B1VAE.safetensors [e5f3cbc5f7] to load dreamshaper_8.safetensors [879db523c3] Dectivating unet: [TRT] ClothingAdjuster3 (realisticVisionV60B1_v60B1VAE) Loading weights [879db523c3] from C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\models\Stable-diffusion\dreamshaper_8.safetensors Loading VAE weights specified in settings: C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\models\VAE\vae-ft-mse-840000-ema-pruned.safetensors Applying attention optimization: xformers... done. Weights loaded in 6.4s (send model to cpu: 0.1s, calculate hash: 0.9s, load weights from disk: 0.3s, apply weights to model: 4.4s, load VAE: 0.1s, move model to device: 0.4s). Activating unet: [TRT] ClothingAdjuster3 (realisticVisionV60B1_v60B1VAE) Loading TensorRT engine: C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\models\Unet-trt\realisticVisionV60B1_v60B1VAE_d6aa28fb_cc75_sample=1x4x64x64+2x4x64x64+8x4x128x128-timesteps=1+2+8-encoder_hidden_states=1x77x768+2x77x768+8x154x768.trt

Loaded Profile: 0 sample = [(1, 4, 64, 64), (2, 4, 64, 64), (8, 4, 128, 128)] timesteps = [(1,), (2,), (8,)] encoder_hidden_states = [(1, 77, 768), (2, 77, 768), (8, 154, 768)] latent = [(-1946027008), (-1946027693), (-1946037504)]

Reusing loaded model dreamshaper_8.safetensors [879db523c3] to load realisticVisionV60B1_v60B1VAE.safetensors [e5f3cbc5f7] Dectivating unet: [TRT] ClothingAdjuster3 (realisticVisionV60B1_v60B1VAE) Loading weights [e5f3cbc5f7] from C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\models\Stable-diffusion\realisticVisionV60B1_v60B1VAE.safetensors Loading VAE weights specified in settings: C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\models\VAE\vae-ft-mse-840000-ema-pruned.safetensors Applying attention optimization: xformers... done. Weights loaded in 2.3s (send model to cpu: 0.1s, calculate hash: 1.0s, load weights from disk: 0.2s, apply weights to model: 0.4s, load VAE: 0.1s, move model to device: 0.5s). Activating unet: [TRT] ClothingAdjuster3 (realisticVisionV60B1_v60B1VAE) Loading TensorRT engine: C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\models\Unet-trt\realisticVisionV60B1_v60B1VAE_d6aa28fb_cc75_sample=1x4x64x64+2x4x64x64+8x4x128x128-timesteps=1+2+8-encoder_hidden_states=1x77x768+2x77x768+8x154x768.trt

Loaded Profile: 0 sample = [(1, 4, 64, 64), (2, 4, 64, 64), (8, 4, 128, 128)] timesteps = [(1,), (2,), (8,)] encoder_hidden_states = [(1, 77, 768), (2, 77, 768), (8, 154, 768)] latent = [(-1946012160), (-1946013440), (-1946010368)]

Restoring base VAE Applying attention optimization: xformers... done. VAE weights loaded. Loading VAE weights specified in settings: C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\models\VAE\vae-ft-mse-840000-ema-pruned.safetensors Applying attention optimization: xformers... done. VAE weights loaded. Dectivating unet: [TRT] ClothingAdjuster3 (realisticVisionV60B1_v60B1VAE) | 0/50 [00:00<?, ?it/s] Activating unet: [TRT] ClothingAdjuster3 (realisticVisionV60B1_v60B1VAE) Loading TensorRT engine: C:\F Backup\AuTomAtic1111\stable-diffusion-webui\stable-diffusion-webui\models\Unet-trt\realisticVisionV60B1_v60B1VAE_d6aa28fb_cc75_sample=1x4x64x64+2x4x64x64+8x4x128x128-timesteps=1+2+8-encoder_hidden_states=1x77x768+2x77x768+8x154x768.trt`

idk it kept looping crazy stuff like that, then like i said, hit refresh icon on the unet drop down one more time and the error was gone and it was going my top 13it/s again =(

[banshek7]

Its actually evolving over time, this is horrifying, I just spent 30 minutes dittling with the fucking pulldown unets and hitting refresh and then trying new args i was so desperate and finally about 9it/s clicked back on. It seems to have a mind of its own, it feels that random, i sit here and its hard as hell to find a repeatable step now to get it to give me full speed. Sometimes it starts at 1it/s now.

Mind you before starting to export other models and loras, i could just spam generate at 13it/s, it seems to be some kind of mismatch between the model and unet or something, i really dont know, i just hope i can fix it or find a bigger clue

[banshek7]

My current and I guess final method for reaching usability now is I load up, now with these args set COMMANDLINE_ARGS=--xformers --force-enable-xformers --theme dark (I took out half vae)

Then my web UI loads with real modelV6, unet real model, & a VAE. If I were to generate its glitched at 1-3 it/s I have to switch my model, set unet to none, and generate, Then I have to go back to my realV6 model, switch unet to automatic, refresh, then go back to my matching Unet TRT model, then with any luck its going at at least 9it/s and 2 it/s upscale, but I have to keep hitting refresh on the unet after each generation (sometimes, sometimes I can just keep generating at full speed)

its beyond retarded, I want to punch it in the fucking face if it had one. I've never been more mad at a technical problem lmfao, because it seems to follow no logic, besides having built in A.I to drive me insane somehow.

I believe the new maximum of about 10it/s after fixing the glitch is the result of taking no half vae out, so when I fix it and hit 10, its the same as when I was hitting 13 before, but the effort to get there is a lot