[llm] Support different shape of input_pos #11869
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
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) [ghstack-poisoned]
larryliu0820
requested review from
JacobSzwejbka,
jackzhxng,
kirklandsign,
lucylq,
mergennachin and
swolchok
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/11869
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 9481d79 with merge base 222d9e3 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
larryliu0820
added a commit
that referenced
this pull request
Jun 24, 2025
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) ghstack-source-id: 292265667 Pull Request resolved: #11869
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D77203700 |
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) [ghstack-poisoned]
larryliu0820
added a commit
that referenced
this pull request
Jun 24, 2025
Pull Request resolved: #11869 For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. ghstack-source-id: 292466463 @exported-using-ghexport Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/)
|
This pull request was exported from Phabricator. Differential Revision: D77203700 |
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) [ghstack-poisoned]
larryliu0820
added a commit
that referenced
this pull request
Jun 24, 2025
Pull Request resolved: #11869 For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. ghstack-source-id: 292469061 @exported-using-ghexport Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/)
|
This pull request was exported from Phabricator. Differential Revision: D77203700 |
kimishpatel
approved these changes
Jun 24, 2025
guangy10
approved these changes
Jun 24, 2025
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) [ghstack-poisoned]
larryliu0820
added a commit
that referenced
this pull request
Jun 25, 2025
Pull Request resolved: #11869 For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. ghstack-source-id: 292517675 @exported-using-ghexport Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/)
|
This pull request was exported from Phabricator. Differential Revision: D77203700 |
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) [ghstack-poisoned]
larryliu0820
added a commit
that referenced
this pull request
Jun 25, 2025
Pull Request resolved: #11869 For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. ghstack-source-id: 292523573 @exported-using-ghexport Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/)
|
This pull request was exported from Phabricator. Differential Revision: D77203700 |
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) [ghstack-poisoned]
larryliu0820
added a commit
that referenced
this pull request
Jun 25, 2025
Pull Request resolved: #11869 For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. ghstack-source-id: 292529628 @exported-using-ghexport Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/)
|
This pull request was exported from Phabricator. Differential Revision: D77203700 |
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) [ghstack-poisoned]
larryliu0820
added a commit
that referenced
this pull request
Jun 25, 2025
Pull Request resolved: #11869 For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. ghstack-source-id: 292529864 @exported-using-ghexport Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/)
|
This pull request was exported from Phabricator. Differential Revision: D77203700 |
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) [ghstack-poisoned]
larryliu0820
added a commit
that referenced
this pull request
Jun 25, 2025
Pull Request resolved: #11869 For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. ghstack-source-id: 292546578 @exported-using-ghexport Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/)
|
This pull request was exported from Phabricator. Differential Revision: D77203700 |
For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/) [ghstack-poisoned]
larryliu0820
added a commit
that referenced
this pull request
Jun 25, 2025
Pull Request resolved: #11869 For huggingface models, `forward()` is taking `tokens` as well as `cache_positions`, which is a list of cache indices. This is different than the .pte files `export_llama` gives, which are taking `tokens` and `input_pos` where `input_pos` is a scalar tensor. This PR adds support inside `text_decoder_runner.cpp` to handle both shapes of `input_pos`/`cache_positions`. To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in `input_pos` or `cache_position`. ghstack-source-id: 292560636 @exported-using-ghexport Differential Revision: [D77203700](https://our.internmc.facebook.com/intern/diff/D77203700/)
|
This pull request was exported from Phabricator. Differential Revision: D77203700 |
facebook-github-bot
merged commit 84c41a8
into
gh/larryliu0820/67/base
102 of 104 checks passed
facebook-github-bot
temporarily deployed
to
cherry-pick-bot
GitHub Actions
Inactive
larryliu0820
added a commit
that referenced
this pull request
Jun 27, 2025
This PR was created by the merge bot to help merge the original PR into the main branch. ghstack PR number: #11869 by @larryliu0820 ^ Please use this as the source of truth for the PR details, comments, and reviews ghstack PR base: https://github.com/pytorch/executorch/tree/gh/larryliu0820/67/base ghstack PR head: https://github.com/pytorch/executorch/tree/gh/larryliu0820/67/head Merge bot PR base: https://github.com/pytorch/executorch/tree/gh/larryliu0820/66/orig Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/larryliu0820/67/orig @diff-train-skip-merge --------- Co-authored-by: Mengwei Liu <[email protected]> Co-authored-by: Mengwei Liu <[email protected]>
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.
Stack from ghstack (oldest at bottom):
For huggingface models,
forward()is takingtokensas well ascache_positions, which is a list of cache indices. This is different than the .pte filesexport_llamagives, which are takingtokensandinput_poswhereinput_posis a scalar tensor.This PR adds support inside
text_decoder_runner.cppto handle both shapes ofinput_pos/cache_positions.To make the logic more generic without relying on extra metadata, here I'm adding the logic of inspecting method meta and input tensor info, to make a decision if we want to feed in
input_posorcache_position.Differential Revision: D77203700