14 Comments

just to point out, dalle mega has been renamed crAIyon and the issue with traffic has solved (hopefully) but I'm also getting errors in the pipeline code, though thankfully that will no longer be an issue as the developers of dalle mini/mega have learned from the lesson and now crAIyon works without the traffic errors

Expand full comment
Jun 16, 2022·edited Jun 17, 2022

so as an update to my issue I decided to pay the piper and buy collab pro, it's working now but I'm not getting near as good predictions as you (the op) and I wanted to know how you got such good results with dall e mega

edit: also want to point out that I'm getting the error "invalid index to scalar variable." on the line print(f"Score: {jnp.asarray(logits[i][idx], dtype=jnp.float32):.2f}\n") and I can't figure out a solution related to the error and this specific line of code

Expand full comment

dead comment section xd

Expand full comment
author

Hey! When you say you aren't getting good results, what exactly do you mean? I think the reason these results look so good is because the cartoony style I was going for is particularly well suited to the model. If you look at my other post in which I review some "higher level" models, that might help a bit. In general I think it is heavily dependent on the style you want. If you're making photorealistic images, you are going to have much rougher results.

Expand full comment

I see, though my main concern was the error I put in after editing my comment, I tried searching the error with collab's built in error search but nothing related to the specific codes

Expand full comment
author

What section/line of the collab is it stopping on? A screenshot would be super useful in troubleshooting!

Expand full comment

This post links to 4 models. The first one is overloaded. The second one is described as "terrifying". I tried the third one and it didn't do well with the prompts I provided. The fourth one ran out of RAM and crashed--looks like the free version doesn't provide enough resources.

Still, this is good to know about. Thanks for doing the research.

Expand full comment

I figured out the issue.

On the second block, it recommends

`# if the notebook crashes too often you can use dalle-mini instead by uncommenting below line

# DALLE_MODEL = "dalle-mini/dalle-mini/mini-1:v0`

Do as it says by removing the # or else you'll run out of RAM and crash without Google Colab Pro. By default it uses the MEGA version which uses far more RAM then the free version allows, telling it to use DALL-E Mini which is what the web app also uses can run under Google Colab free

Expand full comment

can you explain this in greater, step-by-step detail? I'm not a computer guru/geek like the rest of you so I have no idea what exactly to do, I tried removing the # in the line of code below it but that did nothing so I guess I did it wrong

Expand full comment

It should be as simple as simple as removing the # at the start of this line:

# DALLE_MODEL = "dalle-mini/dalle-mini/mini-1:v0"

Then click the play button. You need to click the play button for each code block in order.

Did you get an API key from wandb.ai and enter it when the system prompted you for it?

Expand full comment

I'm still getting an error saying that I'm using too much data, and I tried getting an api key after reading your reply, it didn't make a difference

Expand full comment

Hey, that fixed it. Thank you!

Expand full comment
author

Sorry I'm just now seeing these comments! Dall-e MEGA is a project by the same team that is a real RAM monster. Thank you for the help Dalton!

Expand full comment

thats weird, it didn't use to run out of RAM yesterday but now it's doing it today. looks like I'm not alone?

Expand full comment