Fix duplicate arguments passed to dummy inputs in ONNX export #16045
Conversation
lewtun
changed the title
|
The documentation is not available anymore as the PR was closed or merged. |
lewtun
changed the title
LysandreJik
left a comment
LysandreJik
left a comment
There was a problem hiding this comment.
Looks good! Left two comments that should be applied 4 times each :)
src/transformers/onnx/convert.py
Outdated
| `Tuple[List[str], List[str]]`: A tuple with an ordered list of the model's inputs, and the named inputs from | ||
| the ONNX configuration. | ||
| """ | ||
| from ..tokenization_utils_base import PreTrainedTokenizerBase |
There was a problem hiding this comment.
I think this can be a top-level import
src/transformers/onnx/convert.py
Outdated
| "The `tokenizer` argument is deprecated and will be removed in version 5 of Transformers. Use `preprocessor` instead.", | ||
| FutureWarning, | ||
| ) | ||
| logger.warning("Overwriting the `preprocessor` argument with `tokenizer` to generate dummmy inputs.") |
There was a problem hiding this comment.
Maybe this can be an info as it's more additional information and not really an error
There was a problem hiding this comment.
(warnings get displayed by default, info is displayed when users ask to have more info)
src/transformers/onnx/convert.py
Outdated
| import onnx | ||
| import tf2onnx | ||
|
|
||
| from ..tokenization_utils_base import PreTrainedTokenizerBase |
There was a problem hiding this comment.
Same comment about top-level
src/transformers/onnx/convert.py
Outdated
| "The `tokenizer` argument is deprecated and will be removed in version 5 of Transformers. Use `preprocessor` instead.", | ||
| FutureWarning, | ||
| ) | ||
| logger.warning("Overwriting the `preprocessor` argument with `tokenizer` to generate dummmy inputs.") |
There was a problem hiding this comment.
Same comment about logging level
sgugger
left a comment
sgugger
left a comment
There was a problem hiding this comment.
Good for me with Lysandre's comments! Thanks for working on this!
4 participants
What does this PR do?
This PR fixes:
generate_dummy_inputs()function during the ONNX export.It also removes problematic TensorFlow integration tests, where the model implementation doesn't have parity with the PyTorch one (e.g.
camembert-baseis missing the causal LM head in TensorFlow). I'll address those issues in separate PRs as it involves touching the TensorFlow modeling files.With these fixes, all slow ONNX tests now pass in all environments (only
torch, onlytensorflow,torchandtensorflow):cc @michaelbenayoun