Weekly GitHub Report for Tensorflow: March 24, 2025 - March 31, 2025 (12:10:44)
Weekly GitHub Report for Tensorflow
Thank you for subscribing to our weekly newsletter! Each week, we deliver a comprehensive summary of your GitHub project's latest activity right to your inbox, including an overview of your project's issues, pull requests, contributors, and commit activity.
Table of Contents
I. News
1.1 Recent Version Releases:
The current version of this repository is v2.19.0
1.2 Version Information:
The TensorFlow 2.19.0 release, created on March 5, 2025, introduces breaking changes in the LiteRT (a.k.a. tf.lite) C++ and Python APIs, including the transition of tf.lite.Interpreter to a new location with a deprecation warning. Notable improvements include support for bfloat16 in the tfl.Cast operation, and the discontinuation of publishing libtensorflow packages, although they remain accessible via PyPI.
II. Issues
2.1 Top 5 Active Issues:
We consider active issues to be issues that that have been commented on most frequently within the last week. Bot comments are omitted.
-
TensorFlow on RTX 5090: This issue is about a bug in TensorFlow version 2.20.0.dev20250314, where it fails to work with the NVIDIA RTX 5090 GPU on a Windows 11 system using WSL2 and Ubuntu 22.04.5 LTS. The user has tried various methods, including building from source and using environment scripts, but has been unable to get TensorFlow running, encountering errors related to CUDA compatibility and compute capability discrepancies.
- The comments section includes users sharing similar issues, offering to help, and discussing potential solutions such as using specific versions of LLVM and CUDA. One user documents their attempts to build TensorFlow for the RTX 5090, including a detailed guide and script for others to follow, while others report errors and seek further assistance.
- Number of comments this week: 10
-
Compatibility issue with Python 3.11: This issue involves a compatibility problem with TensorFlow version 2.19.0 when used with Python 3.11.10 on a Windows 10 platform, resulting in a DLL load failure error. The user reports that the error occurs despite having used the same virtual environment and code without issues for the past three years, suggesting a recent change or update might be causing the problem.
- The comments discuss the user's long-term use of a virtual environment and the sudden appearance of the error, with suggestions to check for missing libraries or incompatible system configurations. There is confusion about the TensorFlow version used, as it seems inconsistent with the timeline provided. Additionally, the existence of Python 3.11 three years ago is questioned, indicating potential discrepancies in the user's setup details.
- Number of comments this week: 5
-
can not import tensorflow as tf when set tensorflow =2.15: This issue involves a bug where the user is unable to import TensorFlow as 'tf' when setting the version to 2.15, which prevents the use of a GPU in a Colab notebook environment. The user has attempted to resolve the issue through the Colab community without success and is seeking advice on how to proceed, as using the latest TensorFlow version also causes model training failures.
- The comments discuss the problem of TensorFlow defaulting to CPU usage when the installed version does not match the GPU configuration in Colab. Suggestions include checking the GPU setup and ensuring the environment path is correct, with a code snippet provided to verify GPU availability and set visible devices. The user expresses gratitude for the assistance received.
- Number of comments this week: 5
-
It doesn't support on python3.13: This issue highlights the lack of support for Python 3.13 in TensorFlow version 2.17, which results in installation failures on macOS Sequoia ARM when attempting to install via the terminal. The error message indicates that no compatible TensorFlow distribution is available for Python 3.13, prompting discussions about the need for TensorFlow to align its release cycle with Python's to avoid such compatibility issues.
- The comments discuss the recurring issue of TensorFlow's delayed support for new Python versions, with users expressing frustration over the lack of compatibility with Python 3.13. Some users suggest downgrading Python, while others highlight the importance of supporting the latest Python version due to its adoption in major distributions like Fedora. The conversation also touches on the complexities of TensorFlow's build process and the need for better synchronization with Python's release schedule.
- Number of comments this week: 3
-
TPU not support TensorFlow 2.18 and 2.17.1: This issue reports a bug where importing TensorFlow versions 2.18 and 2.17.1 on a TPU results in a segmentation fault, indicating a critical error that prevents the software from running correctly. The problem persists across different TPU configurations and is suspected to be related to the integration with NumPy 2.0, which was compiled with these TensorFlow versions.
- The comments discuss various attempts to resolve the issue, including trying different TPU configurations and TensorFlow installations, but the segmentation fault persists. Suggestions include ensuring correct TPU setup and using specific TPU images, but these do not resolve the problem. A workaround involving a specific pip install command is mentioned, which avoids the core dump but leads to a new issue where the program gets stuck while attempting to refresh an access token.
- Number of comments this week: 2
2.2 Top 5 Stale Issues:
We consider stale issues to be issues that has had no activity within the last 30 days. The team should work together to get these issues resolved and closed as soon as possible.
- TF-TRT Warning: Could not find TensorRT: This issue involves a user experiencing difficulties with TensorFlow on an Ubuntu 22.04 system, where the TensorRT library cannot be found despite multiple installation attempts. The user, a graduate student, is facing challenges with the NVIDIA driver compatibility for their RTX 3050 TI GPU, which is affecting their ability to work on machine learning projects.
SystemErrorintf.ensure_shapeandtf.compat.v1.ensure_shapewhendtypeofshapeistf.uint64and its value is too large.: This issue pertains to a bug in TensorFlow where usingtf.ensure_shapeortf.compat.v1.ensure_shapewith ashapeofdtypetf.uint64and a value close to 2^64 results in aSystemErrorandOverflowError. The problem has been reproduced with TensorFlow Nightly on a Linux Ubuntu 20.04 system using Python 3.10, and it occurs specifically when the shape value is extremely large, as demonstrated with a provided example.- [DOCS] Missing complex input for Round op: This issue highlights a documentation bug in TensorFlow's
Roundoperation, where the official documentation incorrectly states that a complex tensor can be used as input, but in practice, this does not work as expected. The user reports that they must apply theRoundoperation separately to the real and imaginary parts of the tensor to achieve the desired result, indicating a discrepancy between the documentation and the actual functionality. - tf.raw_ops.Unbatch aborts with "Check failed: d < dims()": This issue involves a bug in TensorFlow version 2.17 where the
tf.raw_ops.Unbatchoperation aborts unexpectedly with an error message "Check failed: d < dims()". The problem has been reproduced using TensorFlow Nightly on a Linux Ubuntu 20.04.3 LTS system with custom code, and it results in an aborted operation and core dump when attempting to unbatch a tensor. - Some
Check Failederrors intf.raw_ops.Unbatch: This issue involves a bug in TensorFlow version 2.15, specifically related totf.raw_ops.Unbatch, where certainCheck Failederrors are causing the program to crash. The problem has been reproduced using TensorFlow Nightly on a Linux Ubuntu 20.04 system with Python 3.10, and detailed error logs indicate mismatches in expected sizes and dimensions during execution.
2.3 Open Issues
This section lists, groups, and then summarizes issues that were created within the last week in the repository.
Issues Opened This Week: 17
Summarized Issues:
- Abortion Errors in TensorFlow 2.20.0-dev20250302: Several issues in TensorFlow version 2.20.0-dev20250302 involve abortion errors that can potentially lead to denial of service attacks. These errors occur in various functions such as
tf.raw_ops.DepthwiseConv2dNativeBackpropInput,tf.nn.max_pool2d,tensorflow.keras.remat,tf.conv, andtf.compat.v2.nn.depthwise_conv2d_backprop_input, often due to invalid index errors or excessively large parameter values.
- Compatibility and Import Errors on Windows: Users have reported compatibility and import errors when running TensorFlow on Windows systems. These issues include ImportErrors due to failed DLL initialization routines and missing dependencies, affecting TensorFlow versions 2.19.0 and 2.9, and causing problems with Python applications built using PyInstaller.
- TensorFlow Installation and Build Issues: Several issues involve problems with installing or building TensorFlow and related applications. These include failures in building Horovod against TensorFlow 2.19 due to missing files, and installation failures on Android and Windows platforms due to missing or incompatible DLLs.
- Memory Leak in TensorFlow Probability: A memory leak has been reported in the
tfp.math.minimizefunction of TensorFlow Probability, where memory usage increases significantly with each run. This issue persists despite attempts to clear sessions and run garbage collection, and is reproducible with both user code and tutorial notebooks.
- Unsupported Padding Option in SeparableConv1D: A user has encountered a ValueError when using the 'causal' padding option with the SeparableConv1D layer in TensorFlow, which was previously supported. This change has led to confusion and concerns about necessary code modifications.
- Training EfficientDet-Lite0 Model on iOS GPU: A user is seeking guidance on training an EfficientDet-Lite0 model using TensorFlow Lite Model Maker without post-processing, as they need to execute post-processing on an iOS GPU. They are facing difficulties with TensorFlow runtime errors during this process.
2.4 Closed Issues
This section lists, groups, and then summarizes issues that were closed within the last week in the repository. This section also links the associated pull requests if applicable.
Issues Closed This Week: 13
Summarized Issues:
- Quantization and Compatibility Issues: The MLIR quantizer in TensorFlow produces asymmetric quantization for int16 activations, which prevents models from running on TFLite Micro with CMSIS-NN due to its requirement for symmetric quantization. Additionally, a user faces challenges compiling a TensorFlow Lite model for the Edge TPU due to dynamic-sized tensors, requiring modifications for compatibility with static-sized tensors and quantization support.
- Graph Execution and Runtime Errors: A Graph execution error occurs during model training on Rocky Linux with TensorFlow 2.18.0, likely due to a mismatch between the runtime and compiled versions of the CuDNN library. Another issue involves a "NotFoundError" on a Colab TPU with TensorFlow 2.8, suggesting an upgrade to the latest version to resolve the problem.
- TensorFlow Model and API Bugs: A bug in TensorFlow 2.x causes numeric instability and gradient explosion when using sparse tensors with
tf.nn.ctc_loss, while dense tensors do not exhibit this issue. Another bug in TensorFlow 2.18 involves thetf.convert_to_tensorAPI failing to assert equality between a tensor and a float value, leading to assertion errors in unit tests.
- Model Conversion and Accuracy Loss: Converting a Keras Hub model to TensorFlow Lite results in significant accuracy loss, particularly with int8 quantization, which is necessary for edge systems. The user seeks assistance in understanding and resolving the accuracy drop after conversion.
- Import and DLL Load Failures: Users encounter ImportErrors and DLL load failures when importing TensorFlow on Windows, with potential causes including missing MSVC redistributables, CPU incompatibility, or incorrect environment setup. Compatibility issues also arise when importing the DeepFace library with Python 3.11.5, resolved by downgrading to Python 3.9.
- Security Vulnerabilities in TensorFlow Operations: Bugs in TensorFlow 2.19.0 involving
tf.raw_ops.Transposeandtf.raw_ops.Unbatchoperations can cause abortion errors, potentially leading to denial of service attacks. These vulnerabilities are demonstrated in provided code snippets.
- Custom Model Saving and Reloading Issues: A user seeks support for saving and reloading a custom TensorFlow model without redefining the class in the inference file, encountering an error stating "Unknown layer 'CategoricalPreprocessor'" during model reloading.
2.5 Issue Discussion Insights
This section will analyze the tone and sentiment of discussions within this project's open and closed issues that occurred within the past week. It aims to identify potentially heated exchanges and to maintain a constructive project environment.
Based on our analysis, there are no instances of toxic discussions in the project's open or closed issues from the past week.
III. Pull Requests
3.1 Open Pull Requests
This section provides a summary of pull requests that were opened in the repository over the past week. The top three pull requests with the highest number of commits are highlighted as 'key' pull requests. Other pull requests are grouped based on similar characteristics for easier analysis. Up to 25 pull requests are displayed in this section, while any remaining pull requests beyond this limit are omitted for brevity.
Pull Requests Opened This Week: 4
Key Open Pull Requests
1. Qualcomm AI Engine Direct - Enable QNN Power Op: This pull request introduces support for the new operation tfl.pow in the Qualcomm AI Engine Direct's QNN, as evidenced by the successful execution of 119 tests across five test suites, and can be reviewed in detail at the provided GitHub URL.
- URL: pull/89860
- Merged: No
- Associated Commits: 554cb
2. Qualcomm AI Engine Direct - Enable BroadcastTo OP: This pull request aims to enable the operation builder for the BroadcastTo operation in the Qualcomm AI Engine Direct and adds two MLIR models to support different data types, as detailed in the commit found at https://github.com/tensorflow/tensorflow/commit/fb50d86e917078d9f24a74b2727f2feaf8f6c531.
- URL: pull/90051
- Merged: No
- Associated Commits: fb50d
3. [TOSA] Fix legalizing CONV bias: This pull request addresses the unification and simplification of the logic for handling bias in various CONV operations within the TOSA framework by consolidating it into a single function, leveraging the TOSA 1.0 specification that allows bias to be of shape [1], and updating the tests to reflect these changes.
- URL: pull/90118
- Merged: No
- Associated Commits: 988a1
Other Open Pull Requests
-
Documentation Link Updates: This pull request focuses on fixing broken documentation links in the
gpu_native.mdfile of the TensorFlow project. The links were updated to point to new LiteRT functional webpage links, ensuring that users can access the correct resources.
3.2 Closed Pull Requests
This section provides a summary of pull requests that were closed in the repository over the past week. The top three pull requests with the highest number of commits are highlighted as 'key' pull requests. Other pull requests are grouped based on similar characteristics for easier analysis. Up to 25 pull requests are displayed in this section, while any remaining pull requests beyond this limit are omitted for brevity.
Pull Requests Closed This Week: 18
Key Closed Pull Requests
1. Fix conv2d dimension mismatch: This pull request addresses a dimension mismatch issue in the tf.nn.conv2d function by removing an unnecessary transpose operation, adjusting the filter shape, and modifying dummy dimensions to ensure compatibility with the 'NHWC' data format, thereby resolving the error and verifying the fix with a test case to ensure the convolution operation executes correctly and produces the expected output shape.
- URL: pull/90150
- Merged: No
- Associated Commits: 70e89, a7d64, 23d23, fc510, 6ea96, f5f75, b336a, 7e8c3, c7df5, 8475d, 097f1, 546b7, 4bd6d
2. [mlir][tosa] Update Tensorflow to match TOSA v1.0 specification (part 4): This pull request updates the TensorFlow project to align with the TOSA v1.0 specification by modifying several operators, including changing attributes to inputs for operations like MatMul and Rescale, updating zero-point handling, and adding support for new rounding modes, thereby addressing broken code and lit tests.
- URL: pull/89907
- Merged: 2025-03-25T16:21:27Z
3. Qualcomm AI Engine Direct - More Op Builders: This pull request introduces several new operation builders for the Qualcomm AI Engine Direct, including FloorDiv, NotEqual, Logistic, Pad, MaxPool2d, Cumsum, GatherNd, Pow, and TransposeConv, along with corresponding unit tests and enhancements to support these operations within the TensorFlow framework, although it was ultimately not merged.
- URL: pull/89727
- Merged: No
Other Closed Pull Requests
- NANOO FP8 support in TensorFlow: This pull request introduces NANOO FP8 support in TensorFlow, referencing related implementations in XLA, JAX, and FLAX. It is documented in a paper available on arXiv, indicating a significant enhancement in TensorFlow's capabilities.
- Time unit correction in TensorFlow: This pull request addresses an issue where the
nodestats::SetScheduledfunction incorrectly handled the scheduled time by treating it as microseconds instead of nanoseconds. The changes adjust the time unit and add a test forStepStatsto ensure accurate time handling.
- Shared local memory optimization for Intel platform: This pull request enables the use of shared local memory for kernel weights on the Intel platform within OpenCL. It results in a compute saving of approximately 6% to 23% for various sizes of a specific model from the Model Zoo.
- Compilation and data type modifications for Qualcomm AI Engine Direct: This pull request modifies TensorFlow to compile QINT16 data types as QUINT16 within the Qualcomm AI Engine Direct. It includes a comprehensive set of tests that all passed successfully, ensuring the changes are robust and reliable.
- Conflict resolution between Abseil and Xlib.h: This pull request resolves a conflict between
absl::Statusin Abseil and the#define Status intmacro in Xlib.h. By undefiningVK_USE_PLATFORM_XLIB_KHR, it excludes the Xlib.h header file from Vulkan.h, resolving compilation errors on specific try bots.
- Removal of unused variable in Qualcomm vendor folder: This pull request involves the removal of an unused variable from the Qualcomm vendor folder in TensorFlow. It successfully passed all 115 tests across 5 test suites, indicating a clean and effective update.
- ROCm build update for TensorFlow: This pull request addresses a minor update for the ROCm build of TensorFlow by replacing the deprecated "pipes.quote" with "shlex.quote". This change resolves a build failure issue, ensuring continued compatibility and functionality.
- Bug fix in TF/TFL to TOSA lowering process: This pull request addresses a bug in the TF/TFL to TOSA lowering process by correcting the rounding behavior of the FloorDiv operation for integer types. It ensures the operation rounds toward negative infinity as intended.
- Typographical error corrections in documentation: This pull request corrects several typographical errors in the documentation strings of the TensorFlow project. It was successfully merged, improving the clarity and accuracy of the project's documentation.
- Unmerged pull requests with spam or irrelevant content: Several pull requests, such as "I am spamming" and "Spamming with an useless file," were not merged due to their inappropriate or irrelevant content. These submissions lacked meaningful contributions to the TensorFlow project.
- Unmerged support for 4-bit quantization in Qualcomm AI Engine Direct: This pull request aimed to introduce support for 4-bit quantization in the Qualcomm AI Engine Direct within TensorFlow. However, it was not merged into the main codebase, indicating potential issues or lack of necessity.
3.3 Pull Request Discussion Insights
This section will analyze the tone and sentiment of discussions within this project's open and closed pull requests that occurred within the past week. It aims to identify potentially heated exchanges and to maintain a constructive project environment.
-
- Toxicity Score: 0.55 (Frustration, accusation of spamming, defensive responses.)
- This GitHub conversation involves username1 expressing concern over the lack of response from the maintainers, while username2 attempts to mediate by suggesting patience. The tone shifts when username3 accuses username1 of spamming, leading to a defensive response from username1. The conversation is marked by frustration and a lack of resolution, with username1 feeling ignored and username3's accusation escalating tensions.
-
- Toxicity Score: 0.55 (Defensive responses, accusations, escalating tension.)
- This GitHub conversation involves username1 expressing concern over the lack of progress on a pull request, while username2 responds with a defensive tone, explaining the challenges faced. The conversation escalates as username1 accuses username2 of not prioritizing the issue, leading to a tense exchange.
IV. Contributors
4.1 Contributors
Active Contributors:
We consider an active contributor in this project to be any contributor who has made at least 1 commit, opened at least 1 issue, created at least 1 pull request, or made more than 2 comments in the last month.
If there are more than 10 active contributors, the list is truncated to the top 10 based on contribution metrics for better clarity.
| Contributor | Commits | Pull Requests | Issues | Comments |
|---|---|---|---|---|
| Venkat6871 | 5 | 5 | 0 | 47 |
| mihaimaruseac | 4 | 0 | 0 | 40 |
| chunhsue | 15 | 8 | 0 | 17 |
| weilhuan-quic | 5 | 2 | 0 | 33 |
| maludwig | 2 | 1 | 1 | 35 |
| gaikwadrahul8 | 3 | 3 | 0 | 20 |
| jiunkaiy | 3 | 3 | 0 | 13 |
| wonjeon | 2 | 6 | 0 | 10 |
| sjh0849 | 0 | 0 | 9 | 8 |
| Jerry-Ge | 1 | 0 | 0 | 13 |