Fix TensorFlow GraphKeys compatibility error for modern TensorFlow 2.x environments #649
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
- Update aiohttp to address potential security vulnerabilities - Maintains compatibility with existing codebase - Addresses dependency security recommendations
…ation - Created comprehensive solution for convenient processed WikiText-103 dataset access - Added two complementary tools: * setup_wikitext103_dataset.py: Lightweight, dependency-free solution * create_processed_wikitext103_dataset.py: Full-featured with WikiGraphs integration - Features: * One-command dataset download and processing * Automatic vocabulary creation with configurable thresholds * Comprehensive validation and integrity checks * Ready-to-use examples and documentation * Cross-platform compatibility - Created WIKITEXT103_SETUP_GUIDE.md with detailed usage instructions - Updated main README.md with quick start section - Addresses 4+ year old Issue google-deepmind#40 from @cp-pc Files added: - wikigraphs/scripts/setup_wikitext103_dataset.py (400+ lines) - wikigraphs/scripts/create_processed_wikitext103_dataset.py (600+ lines) - wikigraphs/WIKITEXT103_SETUP_GUIDE.md (comprehensive guide) - ISSUE_40_SOLUTION.md (GitHub issue response) This solution transforms WikiText-103 setup from complex multi-step process to simple one-command operation, significantly improving researcher productivity.
- Enhanced TensorFlow session isolation between different DM21 functionals - Fixed graph context issues causing all functionals to default to DM21M - Added proper cleanup methods to prevent resource leaks - Added error handling and validation for model loading - Includes comprehensive test suite and documentation Technical changes: - Modified NeuralNumInt.__init__ to explicitly bind sessions to graphs - Added __del__ and close() methods for proper resource management - Enhanced _build_graph with better error handling and validation - Added test script to validate fix and demonstrate correct usage Fixes: google-deepmind#589
- Added comprehensive tensorflow_compatibility.py module with: * Automatic TensorFlow version detection (1.x vs 2.x) * GraphKeys compatibility layer for TF 2.x environments * CleverHans dependency handling including tensorflow-addons mock * Detailed environment diagnostics and troubleshooting guidance - Updated quick_eval_cifar.py to import compatibility layer before CleverHans - Updated requirements.txt with modern TensorFlow 2.x support and dependency guidance - Added comprehensive test_compatibility.py with full test coverage - Added detailed README_ISSUE_46_FIX.md documentation Resolves the 'tensorflow has no attribute GraphKeys' AttributeError that occurs when running unsupervised adversarial training with modern TensorFlow 2.x versions. The solution maintains backward compatibility with TensorFlow 1.15 while enabling researchers to use modern TensorFlow installations with the CleverHans library. Tested on: Python 3.12, TensorFlow 2.18.1, CleverHans 3.1.0
|
Dooray! Failure Notice
Failure Notice
Your message sent to
***@***.***)
has failed to be delivered.
Please refer to the below for details.
* Recipient :
***@***.***)
* Sent time :
2025-08-11T07:10:05
* Subject :
[google-deepmind/deepmind-research] Fix TensorFlow GraphKeys compatibility error for modern TensorFlow 2.x environments (PR #649)
* Remote host said :
Your mail was denied from the receiver.
This message was sent from a notification-only address that cannot accept incoming email.
For more information, please contact ***@***.***
© Dooray!.
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
Resolves Issue #46 where users encountered
AttributeError: module 'tensorflow' has no attribute 'GraphKeys'when running unsupervised adversarial training code with TensorFlow 2.x.Root Cause: TensorFlow 2.x moved
GraphKeysto thetf.compat.v1namespace, but CleverHans library expects the TF 1.x API wheretf.GraphKeyswas directly accessible.Solution Highlights:
tensorflow_compatibility.pywith automatic TF version detection and API bridgingtensorflow-addonsrequirement with intelligent mockingquick_eval_cifar.pyto automatically setup compatibility before CleverHans importsrequirements.txtfor TensorFlow 2.x support while maintaining 1.x backward compatibilitytest_compatibility.py) validating all scenariosTechnical Implementation:
tf.GraphKeys = tf.compat.v1.GraphKeysfor TF 2.xValidation: All tests pass on Python 3.12 + TensorFlow 2.18.1 + CleverHans 3.1.0. The solution enables researchers to reproduce results using modern TensorFlow installations while maintaining compatibility with legacy TF 1.15 setups.
Impact: Fixes a 4-year-old compatibility issue blocking researchers from using modern TensorFlow versions with the unsupervised adversarial training implementation.
Files Modified:
tensorflow_compatibility.py(NEW) - Comprehensive compatibility layerquick_eval_cifar.py(UPDATED) - Integrated compatibility setuprequirements.txt(UPDATED) - Modern TensorFlow supporttest_compatibility.py(NEW) - Full test coverageREADME_ISSUE_46_FIX.md(NEW) - Detailed documentationTesting: Run
python test_compatibility.pyto verify the fix works correctly.