Compliancy with tf-nightly #9570
Conversation
src/transformers/file_utils.py
Outdated
| if _tf_available: | ||
| if version.parse(_tf_version) < version.parse("2"): | ||
| if version.parse(_tf_version) < version.parse("2.3"): |
There was a problem hiding this comment.
No this was just an old check for version 1 that was left, but we should not hardcode anything here, as we will forget to update it when we pin tensorflow to another release.
There was a problem hiding this comment.
Should I let it as it is or completely remove this condition?
There was a problem hiding this comment.
I think we should leave it as is. Let's see what the others think!
There was a problem hiding this comment.
I agree with Sylvain that this should stay as it is, it's to indicate that the general framework detected (TensorFlow 1) will not work correctly as we expect another (TensorFlow 2).
src/transformers/file_utils.py
Outdated
| if _tf_available: | ||
| if version.parse(_tf_version) < version.parse("2"): | ||
| if version.parse(_tf_version) < version.parse("2.3"): |
There was a problem hiding this comment.
I agree with Sylvain that this should stay as it is, it's to indicate that the general framework detected (TensorFlow 1) will not work correctly as we expect another (TensorFlow 2).
|
Ok, just restored the previous version checking. |
| try: | ||
| _tf_version = importlib_metadata.version("tensorflow-gpu") | ||
| except importlib_metadata.PackageNotFoundError: | ||
| try: | ||
| _tf_version = importlib_metadata.version("tf-nightly") | ||
| except importlib_metadata.PackageNotFoundError: | ||
| try: | ||
| _tf_version = importlib_metadata.version("tf-nightly-cpu") | ||
| except importlib_metadata.PackageNotFoundError: | ||
| try: | ||
| _tf_version = importlib_metadata.version("tf-nightly-gpu") | ||
| except importlib_metadata.PackageNotFoundError: | ||
| _tf_version = None | ||
| _tf_available = False |
There was a problem hiding this comment.
Yeah, just done a quick search on pypi to be sure we have all of them 😄
There was a problem hiding this comment.
Might refactor this using a list of names at this stage ;-)
* Compliancy with tf-nightly * Add more version + restore min version check
* Compliancy with tf-nightly * Add more version + restore min version check
What does this PR do?
This PR makes the lib able to be used with the nightly builds of TensorFlow. And fix an issue with the min TensorFlow version.