Load diffusers in native FP16/BF16 precision to reduce the memory usage #1033
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
eaidova
reviewed
Nov 26, 2024
eaidova
reviewed
Nov 26, 2024
eaidova
reviewed
Nov 26, 2024
Co-authored-by: Ekaterina Aidova <ekaterina.aidova@intel.com>
rkazants
reviewed
Nov 26, 2024
Co-authored-by: Roman Kazantsev <roman.kazantsev@intel.com>
eaidova
reviewed
Nov 27, 2024
mvafin
commented
Nov 27, 2024
mvafin
commented
Nov 27, 2024
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
eaidova
approved these changes
Nov 27, 2024
|
@mvafin, any chance to merge it soon? What is the blocker? |
eaidova
requested review from
nikita-savelyevv and
AlexKoff88
and removed request for
rkazants
nikita-savelyevv
approved these changes
Nov 29, 2024
AlexKoff88
reviewed
Nov 29, 2024
| @@ -332,6 +333,48 @@ class StoreAttr(object): | |||
| return model | |||
|
|
|||
| GPTQQuantizer.post_init_model = post_init_model | |||
| elif library_name == "diffusers" and is_safetensors_available() and is_openvino_version(">=", "2024.6"): | |||
| if Path(model_name_or_path).is_dir(): | |||
There was a problem hiding this comment.
Can you please extract and encapsulate this code to a function with a meaningful name?
Only for text generation models. But it even beater there, since we do not use memory to load weights, we just copy them from one location on disk to another using mmap, for diffusers we couldn't achieve this yet. |
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Outdated suggestions cannot be applied.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?
Load diffusers model in native
float16orbfloat16precision and apply 16bit patching for that case.Before submitting