[multimodal] Add token support to MultimodalInput #14451
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,
mergennachin and
swolchok
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/14451
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 7 PendingAs of commit 71dea2a with merge base 07d1092 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
This PR needs a
|
GregoryComer
approved these changes
Sep 19, 2025
kirklandsign
approved these changes
Sep 19, 2025
StrycekSimon
pushed a commit
to nxp-upstream/executorch
that referenced
this pull request
Sep 23, 2025
This pull request adds support for tokenizer-encoded input (as vectors of token IDs) to the `MultimodalInput` class, enabling more flexible and efficient handling of multimodal data. The update includes new constructors, type checks, getters, and factory functions for token inputs, as well as unit tests to ensure correct behavior and compatibility with existing code paths. **MultimodalInput class changes:** * Added a new `TOKENS` type to the `MultimodalInput::Type` enum and updated the internal `std::variant` to support storing `std::vector<uint64_t>` as token data. [[1]](diffhunk://#diff-db31b7448019ab4684675434f5b6e8054ff5d995ffa18e7adee15b5a694a7fb1R34-R73) [[2]](diffhunk://#diff-db31b7448019ab4684675434f5b6e8054ff5d995ffa18e7adee15b5a694a7fb1L290-R367) * Implemented new constructors, type checks (`is_tokens()`), getters (`get_tokens()`), and safe accessors (`try_get_tokens()`) for token inputs, along with static and instance methods for type name conversion. [[1]](diffhunk://#diff-db31b7448019ab4684675434f5b6e8054ff5d995ffa18e7adee15b5a694a7fb1R34-R73) [[2]](diffhunk://#diff-db31b7448019ab4684675434f5b6e8054ff5d995ffa18e7adee15b5a694a7fb1R101-R107) [[3]](diffhunk://#diff-db31b7448019ab4684675434f5b6e8054ff5d995ffa18e7adee15b5a694a7fb1R151-R159) [[4]](diffhunk://#diff-db31b7448019ab4684675434f5b6e8054ff5d995ffa18e7adee15b5a694a7fb1R187-R201) [[5]](diffhunk://#diff-db31b7448019ab4684675434f5b6e8054ff5d995ffa18e7adee15b5a694a7fb1R319-R328) * Added factory functions `make_token_input` for easily creating token-based inputs. **Integration and logging:** * Updated `MultimodalPrefiller::prefill` to handle both text and token inputs, bypassing tokenization when tokens are provided directly. * Added logging in `MultimodalRunner::generate` to include the type name of each input for easier debugging. **Tests:** * Introduced a comprehensive suite of unit tests covering construction, type checking, getters, copy/move semantics, and edge cases for the new token input functionality in `MultimodalInput`.
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.
This pull request adds support for tokenizer-encoded input (as vectors of token IDs) to the
MultimodalInputclass, enabling more flexible and efficient handling of multimodal data. The update includes new constructors, type checks, getters, and factory functions for token inputs, as well as unit tests to ensure correct behavior and compatibility with existing code paths.MultimodalInput class changes:
TOKENStype to theMultimodalInput::Typeenum and updated the internalstd::variantto support storingstd::vector<uint64_t>as token data. [1] [2]is_tokens()), getters (get_tokens()), and safe accessors (try_get_tokens()) for token inputs, along with static and instance methods for type name conversion. [1] [2] [3] [4] [5]make_token_inputfor easily creating token-based inputs.Integration and logging:
MultimodalPrefiller::prefillto handle both text and token inputs, bypassing tokenization when tokens are provided directly.MultimodalRunner::generateto include the type name of each input for easier debugging.Tests:
MultimodalInput.