Learn About GPU Throttling Quota (Another Stupid Guy) :D

Dear Hugging Face Community! :slightly_smiling_face:

First of all, thank you for all your services and for your efforts to democratize artificial intelligence. :sunglasses:

I’m currently fascinated by image generation, especially with models like “DALLE 3 XL v2” and “Fluently XL v4”. I indeed confirm that it is really not safe to use for work. That said, I must admit that as a child I preferred playing with dolls to be with the girls, rather than playing with cars and computers; which partly explains my very limited coding skills. :flushed:

So, after experimenting for free (thanks again) I found myself faced with the message “Error: You have exceeded your GPU quota”. I have partly RTFD (Diffusers, Transformers, Gradio, Datasets, Inference APi, etc. Oh my God, my head is spinning…) and searched the Internet:

I fully understood the principle of limiting GPU allocation and usage.

But, I wrongly understood that creating my own space then switching to the PRO version in order to operate the two image generation models mentioned above, would allow me to have a little more substantial usage than as a free user. However, it really seemed like I was getting the “Error: You have exceeded your GPU quota” error message just as quickly as a free user. :confused:

I fully understand that:
– Image generation consumes a lot of GPU resources
– It is not possible for nine dollars to offer unlimited use

Then, at my very modest level, and due to my current hobby, what is the point of having the PRO version? Except of course, modestly supporting Hugging Face (with great pleasure!) :wink:

If anyone can be kind enough to tell me and explain:
– What is, approximately, the number of 1024x1024px images that can be generated per hour, per 3 hours, or per day. Because quite honestly, this point is very poorly documented, almost non-existent, and it’s very frustrating to click for nothing and wait… :sob:
– What are the models that I can access via API (because it seems that the two models cited above do not allow it) and which will allow me to “Get higher rate limits for serverless inference”.
– Or, which plan to choose to benefit from unlimited use with the two image generation models I mentioned above. Because right now I’m experimenting a lot to find the right prompts and the right combinations of keywords/keyphrases, which results in a lot of images being wasted and unusable by default.

Thank you in advance for your attention.

KissKissBangBang :kissing_heart:

DEVAUX Jean-Charles (the obsessed frenchie) :innocent:

1 Like

Did you ever get any clarity on this?

Unfortunately, that’s been normal, at least for the past few months!
Many people are lamenting the same thing.
I left the request form here.