[Textual inversion] Refactor textual inversion to make it cleaner #5076
Conversation
This file contains hidden or 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
| @@ -2598,7 +2649,6 @@ def load_lora_weights(self, pretrained_model_name_or_path_or_dict: Union[str, Di | |||
|
|
|||
| is_model_cpu_offload = False | |||
| is_sequential_cpu_offload = False | |||
| recursive = False | |||
There was a problem hiding this comment.
No need to introduce a new concept if only used in one function call with recursive=...
patrickvonplaten
changed the title
|
For reviewers: It might be helpful to not just look at the diff, but also check how the new function looks like as a stand-alone file |
sayakpaul
reviewed
Sep 18, 2023
Comment on lines
+971
to
+984
| # 3. Check inputs | ||
| self._check_text_inv_inputs(tokenizer, text_encoder, pretrained_model_name_or_paths, tokens) | ||
|
|
||
| # 4. Load state dicts of textual embeddings | ||
| state_dicts = load_textual_inversion_state_dicts(pretrained_model_name_or_paths, **kwargs) | ||
|
|
||
| # 4. Retrieve tokens and embeddings | ||
| tokens, embeddings = self._retrieve_tokens_and_embeddings(tokens, state_dicts, tokenizer) | ||
|
|
||
| # 5. Extend tokens and embeddings for multi vector | ||
| tokens, embeddings = self._extend_tokens_and_embeddings(tokens, embeddings, tokenizer) | ||
|
|
||
| # 6. Make sure all embeddings have the correct size | ||
| expected_emb_dim = text_encoder.get_input_embeddings().weight.shape[-1] |
sayakpaul
reviewed
Sep 18, 2023
| if len(set(valid_tokens)) < len(valid_tokens): | ||
| raise ValueError(f"You have passed a list of tokens that contains duplicates: {tokens}") | ||
|
|
||
| @staticmethod |
There was a problem hiding this comment.
Just to mark the function as a bit more light weight and allow it to be used just with the class instance
sayakpaul
approved these changes
Sep 18, 2023
yoonseokjin
pushed a commit
to yoonseokjin/diffusers
that referenced
this pull request
Dec 25, 2023
…ggingface#5076) * [Textual inversion] Clean loading * [Textual inversion] Clean loading * [Textual inversion] Clean up * [Textual inversion] Clean up * [Textual inversion] Clean up * [Textual inversion] Clean up
AmericanPresidentJimmyCarter
pushed a commit
to AmericanPresidentJimmyCarter/diffusers
that referenced
this pull request
Apr 26, 2024
…ggingface#5076) * [Textual inversion] Clean loading * [Textual inversion] Clean loading * [Textual inversion] Clean up * [Textual inversion] Clean up * [Textual inversion] Clean up * [Textual inversion] Clean up
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.
What does this PR do?
This PR includes a badly needed refactor of the load_textual_inversion function. The function is now much easier to read & understand. Different concepts are factored out into smaller functions making the overall method much easier to understand.
This PR also makes sure that we remove the model hooks only if we can be sure that the loading will work meaning right before we set the textual inversion embedding. This solves the following issue: #5060