Add optional torch.export.ExportGraphSignature to ONNXProgram #113477
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
When the ONNX model is exported from a torch.export.ExportedProgram, a torch.export.ExportedGraphSignature is available with the specification of the model inputs and outputs. ExportedGraphSignature includes information such as the mapping between the exported input name to the original input/buffer/output name. It also specifies the kind of the input, such as user_input, parameter, buffer or constant_tensor. Outputs kind can be user_output, loss_output, buffer_mutation, etc Such information can be useful to understand what the ONNX model expects as inputs and how the output will look like when the ONNX input/output differs from the original PyTorch input/output schema. When the ONNX model is exported from a Callable or regular torch.nn.MOdule, such information is not available and ONNXProgram.model_signature will yield NOne [ghstack-poisoned]
thiagocrepaldi
requested review from
BowenBao,
abock and
wschin
as code owners
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/113477
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 2c97ecc with merge base 28228e1 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
the
release notes: onnx
This was referenced Nov 10, 2023
thiagocrepaldi
added
module: onnx
…ram" When the ONNX model is exported from a torch.export.ExportedProgram, a torch.export.ExportedGraphSignature is available with the specification of the model inputs and outputs. ExportedGraphSignature includes information such as the mapping between the exported input/buffer/output ONNX name to the original pytorch input/buffer/output name. It also specifies the kind of the input, such as user_input, parameter, buffer or constant_tensor. Outputs kind can be user_output, loss_output, buffer_mutation, etc Such information can be useful to understand what the ONNX model expects as inputs and how the output will look like when the ONNX input/output differs from the original PyTorch input/output schema. When the ONNX model is exported from a Callable or regular torch.nn.MOdule, such information is not available and ONNXProgram.model_signature will yield NOne [ghstack-poisoned]
thiagocrepaldi
added
the
triaged
…ram" When the ONNX model is exported from a torch.export.ExportedProgram, a torch.export.ExportedGraphSignature is available with the specification of the model inputs and outputs. ExportedGraphSignature includes information such as the mapping between the exported input/buffer/output ONNX name to the original pytorch input/buffer/output name. It also specifies the kind of the input, such as user_input, parameter, buffer or constant_tensor. Outputs kind can be user_output, loss_output, buffer_mutation, etc Such information can be useful to understand what the ONNX model expects as inputs and how the output will look like when the ONNX input/output differs from the original PyTorch input/output schema. When the ONNX model is exported from a Callable or regular torch.nn.MOdule, such information is not available and ONNXProgram.model_signature will yield NOne [ghstack-poisoned]
added 3 commits
November 14, 2023 22:02
…ram" When the ONNX model is exported from a torch.export.ExportedProgram, a torch.export.ExportedGraphSignature is available with the specification of the model inputs and outputs. ExportedGraphSignature includes information such as the mapping between the exported input/buffer/output ONNX name to the original pytorch input/buffer/output name. It also specifies the kind of the input, such as user_input, parameter, buffer or constant_tensor. Outputs kind can be user_output, loss_output, buffer_mutation, etc Such information can be useful to understand what the ONNX model expects as inputs and how the output will look like when the ONNX input/output differs from the original PyTorch input/output schema. When the ONNX model is exported from a Callable or regular torch.nn.MOdule, such information is not available and ONNXProgram.model_signature will yield NOne [ghstack-poisoned]
…ram" When the ONNX model is exported from a torch.export.ExportedProgram, a torch.export.ExportedGraphSignature is available with the specification of the model inputs and outputs. ExportedGraphSignature includes information such as the mapping between the exported input/buffer/output ONNX name to the original pytorch input/buffer/output name. It also specifies the kind of the input, such as user_input, parameter, buffer or constant_tensor. Outputs kind can be user_output, loss_output, buffer_mutation, etc Such information can be useful to understand what the ONNX model expects as inputs and how the output will look like when the ONNX input/output differs from the original PyTorch input/output schema. When the ONNX model is exported from a Callable or regular torch.nn.MOdule, such information is not available and ONNXProgram.model_signature will yield NOne [ghstack-poisoned]
…ram" When the ONNX model is exported from a torch.export.ExportedProgram, a torch.export.ExportedGraphSignature is available with the specification of the model inputs and outputs. ExportedGraphSignature includes information such as the mapping between the exported input/buffer/output ONNX name to the original pytorch input/buffer/output name. It also specifies the kind of the input, such as user_input, parameter, buffer or constant_tensor. Outputs kind can be user_output, loss_output, buffer_mutation, etc Such information can be useful to understand what the ONNX model expects as inputs and how the output will look like when the ONNX input/output differs from the original PyTorch input/output schema. When the ONNX model is exported from a Callable or regular torch.nn.MOdule, such information is not available and ONNXProgram.model_signature will yield NOne [ghstack-poisoned]
|
@BowenBao PTAL |
BowenBao
reviewed
Nov 15, 2023
| ) | ||
| """ | ||
|
|
||
| return self._model_signature |
There was a problem hiding this comment.
I feel this is only accurate up till the point for fx-graph in ExportedProgram. Further process inside our exporter that modifies the signature will make this outdated. Unless we apply similar io adapting to this too.
There was a problem hiding this comment.
when user model is torch.export.ExportedProgram, we don't have any pass that changes the signature.
The only pass applied to the resulting graph module is fx_module = passes.InsertTypePromotion(diagnostic_context, fx_module).run().
The model signature is only available for this scenario
There was a problem hiding this comment.
I will update the model_signature/io_adapter docstrings to make this scenario explicit
If a pass alters input/output then the module_signature has to be updated by a input/output adapter. That is a good point for a future pass (hopefully never happens)
BowenBao
approved these changes
Nov 16, 2023
…ram" When the ONNX model is exported from a torch.export.ExportedProgram, a torch.export.ExportedGraphSignature is available with the specification of the model inputs and outputs. ExportedGraphSignature includes information such as the mapping between the exported input/buffer/output ONNX name to the original pytorch input/buffer/output name. It also specifies the kind of the input, such as user_input, parameter, buffer or constant_tensor. Outputs kind can be user_output, loss_output, buffer_mutation, etc Such information can be useful to understand what the ONNX model expects as inputs and how the output will look like when the ONNX input/output differs from the original PyTorch input/output schema. When the ONNX model is exported from a Callable or regular torch.nn.MOdule, such information is not available and ONNXProgram.model_signature will yield NOne [ghstack-poisoned]
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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):
When the ONNX model is exported from a torch.export.ExportedProgram, a
torch.export.ExportedGraphSignature is available with the specification
of the model inputs and outputs.
ExportedGraphSignature includes information such as the mapping between
the exported input/buffer/output ONNX name to the original pytorch input/buffer/output name.
It also specifies the kind of the input, such as user_input, parameter,
buffer or constant_tensor. Outputs kind can be user_output, loss_output,
buffer_mutation, etc
Such information can be useful to understand what the ONNX model expects
as inputs and how the output will look like when the ONNX input/output
differs from the original PyTorch input/output schema.
When the ONNX model is exported from a Callable or regular
torch.nn.MOdule, such information is not available and
ONNXProgram.model_signature will yield NOne