[multimodal] Allow float32 image input #14359
Merged
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
larryliu0820
requested review from
jackzhxng,
lucylq,
mergennachin and
swolchok
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/14359
Note: Links to docs will display an error until the docs builds have been completed. ⏳ No Failures, 4 PendingAs of commit 4d6d3be with merge base d25c35a ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
kirklandsign
approved these changes
Sep 16, 2025
jackzhxng
reviewed
Sep 17, 2025
Contributor
jackzhxng
left a comment
jackzhxng
left a comment
There was a problem hiding this comment.
nice. this would be a good opportunity to get started on a multimodal_prefiller test file too
|
|
||
| executorch::runtime::Result<executorch::extension::TensorPtr> toTensor( | ||
| bool with_batch = false) const { | ||
| // Note: This creates a 3D tensor (CHW). The model might expect a 4D |
Contributor
There was a problem hiding this comment.
seems like you already batch using with_batch so can rm this comment?
Contributor
Author
There was a problem hiding this comment.
I wish it's easy to test. My plan is to setup some python test in the pybind PR
Summary: Test Plan: Reviewers: Subscribers: Tasks: Tags:
larryliu0820
force-pushed
the
image_takes_float
branch
from
4f68739 to
4d6d3be
Compare
Contributor
|
Cherry-pick candidate? |
Contributor
Author
could be. I have a few other PRs so let me land everything and see |
Contributor
Author
|
@pytorchbot cherry-pick --onto release/1.0 -c critical |
pytorchbot
pushed a commit
that referenced
this pull request
Sep 22, 2025
Letting `Image` class support both `uint8_t` and `float` data types, changing `MultimodalPrefiller` class to support text, image, and audio modalities with error checking and modularity. **Image Data Handling and Type Safety:** * Refactored the `Image` class in `image.h` from a simple struct to a class that uses a `std::variant` to support both `uint8_t` and `float` image data, providing type-safe accessors and a `toTensor` method for conversion to tensors. * Updated `load_image` in Llava `main.cpp` to construct `Image` objects using the new class interface and move semantics, ensuring correct data layout and encapsulation. * Added a runtime check in `LlavaImagePrefiller` to ensure only `uint8_t` images are processed, using the new type-checking methods. **Multimodal Prefill Logic and Flexibility:** * Updated the `MultimodalPrefiller` class in `multimodal_prefiller.h` to dynamically check input types, validate tensor types against model expectations, and handles encoder/decoder execution with improved error handling and modularity. (cherry picked from commit bc18834)
Collaborator
Cherry picking #14359The cherry pick PR is at #14490 and it is recommended to link a critical cherry pick PR with an issue. The following tracker issues are updated: Details for Dev Infra teamRaised by workflow job |
StrycekSimon
pushed a commit
to nxp-upstream/executorch
that referenced
this pull request
Sep 23, 2025
Letting `Image` class support both `uint8_t` and `float` data types, changing `MultimodalPrefiller` class to support text, image, and audio modalities with error checking and modularity. **Image Data Handling and Type Safety:** * Refactored the `Image` class in `image.h` from a simple struct to a class that uses a `std::variant` to support both `uint8_t` and `float` image data, providing type-safe accessors and a `toTensor` method for conversion to tensors. * Updated `load_image` in Llava `main.cpp` to construct `Image` objects using the new class interface and move semantics, ensuring correct data layout and encapsulation. * Added a runtime check in `LlavaImagePrefiller` to ensure only `uint8_t` images are processed, using the new type-checking methods. **Multimodal Prefill Logic and Flexibility:** * Updated the `MultimodalPrefiller` class in `multimodal_prefiller.h` to dynamically check input types, validate tensor types against model expectations, and handles encoder/decoder execution with improved error handling and modularity.
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.
Letting
Imageclass support bothuint8_tandfloatdata types, changingMultimodalPrefillerclass to support text, image, and audio modalities with error checking and modularity.Image Data Handling and Type Safety:
Imageclass inimage.hfrom a simple struct to a class that uses astd::variantto support bothuint8_tandfloatimage data, providing type-safe accessors and atoTensormethod for conversion to tensors.load_imagein Llavamain.cppto constructImageobjects using the new class interface and move semantics, ensuring correct data layout and encapsulation.LlavaImagePrefillerto ensure onlyuint8_timages are processed, using the new type-checking methods.Multimodal Prefill Logic and Flexibility:
MultimodalPrefillerclass inmultimodal_prefiller.hto dynamically check input types, validate tensor types against model expectations, and handles encoder/decoder execution with improved error handling and modularity.