If you'd like to support me and do more:
https://ko-fi.com/edgonkofi - commissions open!
If you're looking for a >>LoRA Making Tutorial<<
Summer Dress Code - by EDG
It's getting hot, but it's no reason to not stay fashionable.
Description
FAQ
Comments (28)
tl;dr Would you have any idea why I have dozens of "lora_te1_text_model_encoder_layers" in my console output?
So I noticed weird console output when I use your Fishnet&Shorts LoRA and this LoRA. I'm getting a couple dozen lines that all start with "lora_te1_text_model_encoder_layers". I don't get this behavior will all of your LoRAs (I tested European Vintage and I got my typical console outputs).
I'm pretty sure I'm still actually loading your LoRA too (unless someone recognizes that console log output and knows otherwise). I tested two identical seeds with the exact same prompts (including the trigger word) with and without your LoRA and got different images.
I'm mostly just wondering if you've done anything different between your LoRAs since you've come back and your older LoRAs. Any insight you can provide would be helpful since I just want to know if the weirdness is on my end or just something new you're doing.
same issue for me too, Automatic1111 v. 1.3.2
@TopazStudio I'm not sure it's an error or an issue. I copied the whole console output (~8600 lines), and there's no mention of "error" or "exception" anywhere in there. I think the LoRA works, it's just doing something strange.
The only thing I have done was updating ss kohya, the training tool.
I thought it wouldn't create any issues, and yet, here we are...
I am sincerely sorry
@EDG I guess I was still unclear. The LoRA still works. I still made images. In fact I have about half the images I want to use for my review ready. The only problem is that my console output gets spammed. I don't know if the thing that's causing this is on your end or our end.
The only problem I can see is that a lot of people may see the console log and assume that the LoRA didn't work. I can't speak for everyone else, but I was able to use it just fine.
I should also mention that I run an older version of the webui. For whatever reason, newer versions are unbearably laggy even though my computer is pretty good. It could just be a weird conflict between the new version you're using to train your LoRAs and my outdated webui.
If you think you need any version information from me, don't hesitate to let me know. I made a pastebin of my console output too if you think that might help you, though it looks like I might have a different thing going on than other people.
I've found a solution
@neilarmstron12 @EDG yes, the block at 8653 in the pastebin is the same thing I got too. The LORA is loading and working for me despite the messages, I can get very similar results to the sample images if i follow the same prompts
There seems to be some errors while reading the model.
@_LJ_ I don't know if you saw my own comment, but if you're having the same console output as me, I didn't see any mention of "error" or "expcetion" in the logs. I'm personally still able to make images and the LoRA is having an effect, I just get like 8000 lines of text in my console whenever I use this LoRA.
Hello
It looks like I shouldn't have updated the trainer...
@_LJ_ does the LoRA produce images for you or not?
@EDG yes, it can create the image
@neilarmstron12 there's appear the same massage in console log
@_LJ_ We might have different console outputs then. I was able to determine that the LoRA still works for me by generating two images with the same seed. I kept all prompts the same (even the activation word) and just had one image with the LoRA on and the other one with the LoRA off. If you have two different images and the one with the LoRA on appears to match EDG's example images then the LoRA is working for you. If that's the case, you're just getting the weird console spam that I'm getting.
Otherwise.... I dunno what to do. Hopefully EDG can fix this.
@neilarmstron12 F I X E D
It aggravates me immensely that many of you are not able to run the latest releases without issue.
Let me change the Kohya commit I'm on, the only thing I have changed was the LoRA trainer, so it must be the problem.
I get a "failed to match keys when loading lora" in console but the lora does seem to be somewhat working
You made it work beautifully on your review, thank you very much! <3
It must be a kohya issue, I need to find a commit that doesn't cause any problems.
@EDG thank you for your loras and hard work!
I've found a fix
@EDG awesome! redownloading!
@feetie Please let me know if it's good for you, I've tested locally and with a friend, it was OK. How about you?
@EDG Just tested, no more errors in console, works perfectly!
@feetie YES!!
I have retrained the LoRA, and have had no issue. I've uploaded the new file under the same version.
FIXED!
This should be called "really fkin hot dress generator" !
Getting some amazing results. Thanks
Bought you a coffee
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.