Added the 4k model #5
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I noticed the 4K model hadn't been included yet, so I opted to make the necessary changes. A couple of things worth mentioning: I wasn't sure whether the P1 in SanaMS_1600M_P1_D20 represented patch size, pe interpolation, or something else entirely so I just left it as it was in the others. Additionally, my desire to be helpful didn't quite hold out long enough to manually change the resolution scaling numbers, and so I just left it to Claude 3.5 Sonnet. Obviously LLMs are not especially well suited to math, and while a cursory inspection of the numbers looks solid, you should probably double check them.
Edit: Hmm, so actually it would appear that all is not well after all. Before making this PR, I only did a single test generation with the changes, and that particular combination of seed/prompt was apparently just a fluke, because every image I've been able to generate since has ranged from subpar to unholy abomination. I'm still testing some things, and will update back if I have an luck.
Edit x 2: After a bit more testing, while I think it definitely could use some tweaking, a lot of my issues stemmed from poor Sana prompting practices due to my lack of experience with the model. After making some changes to my prompts, I've been able to get results of reasonable quality.