[ET-VK] Handle scalar tensor and mutable buffer inputs in Vulkan delegate runtime #5930
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
…gate runtime ## Context * Handle scalar tensor inputs by adding them to the graph as symbolic ints * Add support for symint inputs in the Vulkan delegate * Add type checking for Vulkan delegate inputs and outputs This is needed for Transformer models, which receive a an `input_pos` integer scalar tensor as an input. `input_pos` is used in KV cache updates and determines the sizes of the cache slices. ### Why are scalar tensors added as symint? Adding scalar tensors as symints makes more sense than adding them as real tensors, since symints are commonly used to inform tensor shapes. Adding scalar tensors as symints allow them to be easily accessible by the CPU at graph encoding and resizing time, as well as easily accesible by the GPU within compute shaders. Differential Revision: [D63979312](https://our.internmc.facebook.com/intern/diff/D63979312/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/5930
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New FailureAs of commit e34f3f5 with merge base aad548c ( NEW FAILURE - The following job has failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
ciflow/periodic
module: vulkan
facebook-github-bot
added
the
CLA Signed
This was referenced Oct 7, 2024
|
This pull request was exported from Phabricator. Differential Revision: D63979312 |
SS-JIA
added a commit
that referenced
this pull request
Oct 7, 2024
…gate runtime ## Context * Handle scalar tensor inputs by adding them to the graph as symbolic ints * Add support for symint inputs in the Vulkan delegate * Add type checking for Vulkan delegate inputs and outputs This is needed for Transformer models, which receive a an `input_pos` integer scalar tensor as an input. `input_pos` is used in KV cache updates and determines the sizes of the cache slices. ### Why are scalar tensors added as symint? Adding scalar tensors as symints makes more sense than adding them as real tensors, since symints are commonly used to inform tensor shapes. Adding scalar tensors as symints allow them to be easily accessible by the CPU at graph encoding and resizing time, as well as easily accesible by the GPU within compute shaders. Differential Revision: [D63979312](https://our.internmc.facebook.com/intern/diff/D63979312/) ghstack-source-id: 246578362 Pull Request resolved: #5930
…Vulkan delegate runtime" ## Context * Handle scalar tensor inputs by adding them to the graph as symbolic ints * Add support for symint inputs in the Vulkan delegate * Add type checking for Vulkan delegate inputs and outputs This is needed for Transformer models, which receive a an `input_pos` integer scalar tensor as an input. `input_pos` is used in KV cache updates and determines the sizes of the cache slices. ### Why are scalar tensors added as symint? Adding scalar tensors as symints makes more sense than adding them as real tensors, since symints are commonly used to inform tensor shapes. Adding scalar tensors as symints allow them to be easily accessible by the CPU at graph encoding and resizing time, as well as easily accesible by the GPU within compute shaders. Differential Revision: [D63979312](https://our.internmc.facebook.com/intern/diff/D63979312/) [ghstack-poisoned]
SS-JIA
added a commit
that referenced
this pull request
Oct 7, 2024
…gate runtime Pull Request resolved: #5930 ## Context * Handle scalar tensor inputs by adding them to the graph as symbolic ints * Add support for symint inputs in the Vulkan delegate * Add type checking for Vulkan delegate inputs and outputs This is needed for Transformer models, which receive a an `input_pos` integer scalar tensor as an input. `input_pos` is used in KV cache updates and determines the sizes of the cache slices. Additionally, mutable buffer inputs/outputs, which appear as `TensorRef` to the Vulkan graph, are handled as well by ignoring them when copying outputs. More details in the comments. ### Why are scalar tensors added as symint? Adding scalar tensors as symints makes more sense than adding them as real tensors, since symints are commonly used to inform tensor shapes. Adding scalar tensors as symints allow them to be easily accessible by the CPU at graph encoding and resizing time, as well as easily accesible by the GPU within compute shaders. Differential Revision: [D63979312](https://our.internmc.facebook.com/intern/diff/D63979312/) ghstack-source-id: 246588007
|
This pull request was exported from Phabricator. Differential Revision: D63979312 |
…Vulkan delegate runtime" ## Context * Handle scalar tensor inputs by adding them to the graph as symbolic ints * Add support for symint inputs in the Vulkan delegate * Add type checking for Vulkan delegate inputs and outputs This is needed for Transformer models, which receive a an `input_pos` integer scalar tensor as an input. `input_pos` is used in KV cache updates and determines the sizes of the cache slices. ### Why are scalar tensors added as symint? Adding scalar tensors as symints makes more sense than adding them as real tensors, since symints are commonly used to inform tensor shapes. Adding scalar tensors as symints allow them to be easily accessible by the CPU at graph encoding and resizing time, as well as easily accesible by the GPU within compute shaders. Differential Revision: [D63979312](https://our.internmc.facebook.com/intern/diff/D63979312/) [ghstack-poisoned]
|
This pull request was exported from Phabricator. Differential Revision: D63979312 |
SS-JIA
added a commit
that referenced
this pull request
Oct 7, 2024
…gate runtime Pull Request resolved: #5930 ## Context * Handle scalar tensor inputs by adding them to the graph as symbolic ints * Add support for symint inputs in the Vulkan delegate * Add type checking for Vulkan delegate inputs and outputs This is needed for Transformer models, which receive a an `input_pos` integer scalar tensor as an input. `input_pos` is used in KV cache updates and determines the sizes of the cache slices. Additionally, mutable buffer inputs/outputs, which appear as `TensorRef` to the Vulkan graph, are handled as well by ignoring them when copying outputs. More details in the comments. ### Why are scalar tensors added as symint? Adding scalar tensors as symints makes more sense than adding them as real tensors, since symints are commonly used to inform tensor shapes. Adding scalar tensors as symints allow them to be easily accessible by the CPU at graph encoding and resizing time, as well as easily accesible by the GPU within compute shaders. ghstack-source-id: 246627218 Differential Revision: [D63979312](https://our.internmc.facebook.com/intern/diff/D63979312/)
jorgep31415
approved these changes
Oct 7, 2024
…Vulkan delegate runtime" ## Context * Handle scalar tensor inputs by adding them to the graph as symbolic ints * Add support for symint inputs in the Vulkan delegate * Add type checking for Vulkan delegate inputs and outputs This is needed for Transformer models, which receive a an `input_pos` integer scalar tensor as an input. `input_pos` is used in KV cache updates and determines the sizes of the cache slices. ### Why are scalar tensors added as symint? Adding scalar tensors as symints makes more sense than adding them as real tensors, since symints are commonly used to inform tensor shapes. Adding scalar tensors as symints allow them to be easily accessible by the CPU at graph encoding and resizing time, as well as easily accesible by the GPU within compute shaders. Differential Revision: [D63979312](https://our.internmc.facebook.com/intern/diff/D63979312/) [ghstack-poisoned]
SS-JIA
added a commit
that referenced
this pull request
Oct 8, 2024
…gate runtime Pull Request resolved: #5930 ## Context * Handle scalar tensor inputs by adding them to the graph as symbolic ints * Add support for symint inputs in the Vulkan delegate * Add type checking for Vulkan delegate inputs and outputs This is needed for Transformer models, which receive a an `input_pos` integer scalar tensor as an input. `input_pos` is used in KV cache updates and determines the sizes of the cache slices. Additionally, mutable buffer inputs/outputs, which appear as `TensorRef` to the Vulkan graph, are handled as well by ignoring them when copying outputs. More details in the comments. ### Why are scalar tensors added as symint? Adding scalar tensors as symints makes more sense than adding them as real tensors, since symints are commonly used to inform tensor shapes. Adding scalar tensors as symints allow them to be easily accessible by the CPU at graph encoding and resizing time, as well as easily accesible by the GPU within compute shaders. ghstack-source-id: 246752221 Differential Revision: [D63979312](https://our.internmc.facebook.com/intern/diff/D63979312/)
|
This pull request was exported from Phabricator. Differential Revision: D63979312 |
|
This pull request has been merged in 62a13c1. |
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):
higher_order_auto_functionalized#5884Context
This is needed for Transformer models, which receive a an
input_posinteger scalar tensor as an input.input_posis used in KV cache updates and determines the sizes of the cache slices.Why are scalar tensors added as symint?
Adding scalar tensors as symints makes more sense than adding them as real tensors, since symints are commonly used to inform tensor shapes. Adding scalar tensors as symints allow them to be easily accessible by the CPU at graph encoding and resizing time, as well as easily accesible by the GPU within compute shaders.
Differential Revision: D63979312