[DRAFT] Support multiple tokenizers and other layers with assets #1860
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
Preset saving and loading does not currently generalize to multiple
tokenizers (or other preprocessor with static assets), this is a work
in progress PR towards adding it, specifically for stable diffusion.
The high-level api would allow something like this
```python
# High-level loading.
image_to_text = keras_hub.models.ImageToText.from_preset(
"sd3_preset",
)
# Low-level tokenizer loading.
clip_l_tokenizer = kersa_hub.tokenizers.Tokenizer.from_preset(
"sd3_preset", config_file="clip_l_tokenizer.json",
)
clip_g_tokenizer = kersa_hub.tokenizers.Tokenizer.from_preset(
"sd3_preset", config_file="clip_g_tokenizer.json",
)
```
During conversion, we would need to make sure each tokenizer was
created with a separate `config_file` passed to the constructor. Then
when calling `task.save_to_preset("path")`, you would get the following
structure.
```
assets/clip_l_tokenizer/...
assets/clip_g_tokenizer/...
assets/t5_tokenizer/...
clip_l_tokenizer.json
clip_g_tokenizer.json
t5_tokenizer.json
```
|
@james77777778 just creating this to share a possible solution to #1820 (comment), feel free to patch in and play around. Though I would still merge #1820 without a solution here so we can stay incremental. |
|
I will try this PR when porting the weights of SD3. |
Hi James! Please feel free to take this PR up. |
|
Yes sorry for the delay, but please take this PR up. Let's test this out with sd3 |
1 similar comment
|
Yes sorry for the delay, but please take this PR up. Let's test this out with sd3 |
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Outdated suggestions cannot be applied.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.
Preset saving and loading does not currently generalize to multiple tokenizers (or other preprocessor with static assets), this is a work in progress PR towards adding it, specifically for stable diffusion.
The high-level api would allow something like this
During conversion, we would need to make sure each tokenizer was created with a separate
config_filepassed to the constructor. Then when callingtask.save_to_preset("path"), you would get the following structure.This is just a WIP commit to share with @james77777778 , tests will not pass yet.