Runtime error

#26
by Arnon2 - opened

I can't seem to run this space, it says "Runtime error on A10G". It was fine a few days ago... Can you please check it up?

It seems a lot of spaces got taken down recently for abusing storage and/or processing limits, and there's a bug where it tries to rebuild and either eventually produces a runtime error or freezes altogether. You can try a factory reset or cloning the space and relaunching it from there. @dvruette

It seems to me the problem lies in outdated requirements. I somehow forced it to run on my copy by tampring around, but it's way too slow without cashing out, and I don't have free money as it it... I really hope nothing happened to author, and he can eventually patch this thing up.

Arnon2 changed discussion status to closed
Arnon2 changed discussion status to open

@Arnon2 !

I want help and found that is :

It is an accelerate outdated package version.
But even upgrading that to the 0.20.3 version.. It will not work (but initial quick build error disappeared) and fall under another error showing something weird :

text_model.load_state_dict(text_model_dict)
File "/home/user/.local/lib/python3.10/site-packages/torch/nn/modules/module.py", line 2041, in load_state_dict
raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for CLIPTextModel:
Unexpected key(s) in state_dict: "text_model.embeddings.position_ids"

Unfortunately..
I was unable to go further after that error.

@FashionStash As I said... I resolved an error... but I'm not an author of This, therefore, I'm not sure, if I can actually fix it. And even if I host myself, I hove no money to run it at full speed, as author did...

Oh.. Oh! @Arnon2
But.. Just for I can understanding you well..
Even with a free plan, here at huggingface.co, you was able to make it work?

If yes.. Then..
If you find out how to get working, let the community know how!

Perhaps some owner of paid Spaces would be able to duplicate !

Because... Myself.. I just resolved the requirements error not the other one related to CLIPTextModel.

I wasn't unlucky to find out all the right dependencies, trying during hours.

But even from that side..
I think the author must fix the fabric GITHUB itself where that app import FABRIC repo.

Sidenote : dvruette fabric GITHUB repo seems last updated 3 weeks ago.. So that let us some hope 🙏😅

Good luck @dvruette !

Yes this should be an easy fix, just need to fix the package versions to what they were before it broke. Going to take a look at this tomorrow.

Finally, I can comment again, and yay, @dvruette is here! By the way, to make things slightly easier, that's what I did.
1)As it's known, I changed accelerate>=0.20.3.
2)I also did transformers==4.30.2, but, I don't think it did anything.
3)I also upgraded DreamShaper to 8 in app.py. It's probably unrelated, but still...
4)This was the step, that made space working, but... now that I'm thinking about it it kind looks like amateur-ish anti-theft measure, so I'm not sure, is it's appropriate to mention it here... it's related to dreamlike photoreal... but that's probably what you mean by package versions (I don't really know much about how this entire thing work, now).
Oh well, it was knida fun, figuring out how to make it operational, but now I can get to generating pictures normally.

Running again, fixing the transformers version seems to have worked.

dvruette changed discussion status to closed

@dvruette , it's crashed again... I think space is just stuck, and needs a reboot, but it may be that DreamShaper7 expired. Either way, it requires you to restart it. many thanks in advance.

Arnon2 changed discussion status to open

I’ve restarted, let’s see if it helps.

Thank you.

Arnon2 changed discussion status to closed

@dvruette , It's me again. I'm sorry to bother, but HuggungFace was down, and now space requires a reboot. Can you boot it up please? Many thanks in advance.

Arnon2 changed discussion status to open

Thanks for the ping! I've restarted it, let me know if it stops working again.

@dvruette , sure thing, I'm not going from here nowhere soon. it's very good thing, pretty much a miniature AI training program. Many thanks again.

Arnon2 changed discussion status to closed

Sign up or log in to comment