Weekly GitHub Report for Tensorflow: May 19, 2025 - May 26, 2025 (12:00:28)
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 to the LiteRT C++ and Python APIs, including the transition of tf.lite.Interpreter to a new location, with a deprecation warning for its future removal in version 2.20. Additionally, the update enhances tf.lite by supporting bfloat16 in the tfl.Cast operation and discontinues publishing libtensorflow packages, though 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.
-
Segmentation fault in tf.repeat: This issue reports a segmentation fault occurring when using the
tf.repeatfunction in TensorFlow version 2.19.0 on Ubuntu 18.04 with Python 3.11.11, which is suspected to be related to the use of deeply nested tuples in the input. The problem has been reproduced with TensorFlow Nightly, and the user has provided a standalone code snippet to demonstrate the error.- The comments discuss the possibility of an out-of-memory (OOM) issue due to deeply nested tuples, with a suggestion to simplify the input structure. The user clarifies that the input was generated through fuzzing, and a recommendation is made to triage fuzzing reports before filing bugs, as this case seems to be an OOM rather than a security issue. It is advised to compile with ASAN and provide a stack trace for better analysis.
- Number of comments this week: 3
-
Question about fuzzing TensorFlow with Atheris(oss-fuzz): This issue is about a user inquiring whether the entire TensorFlow project can be compiled with the
-fsanitize=fuzzer-no-linkflag to collect C++ coverage using Atheris, as part of the fuzzing efforts in the OSS-Fuzz tensorflow-py project. The user is seeking clarification on why this approach is not currently implemented in the project and whether it would improve the efficiency of fuzzing.- The first comment asks the user to elaborate on the feature and specify its use cases. The second comment suggests that while the approach might be beneficial, the TensorFlow library's large size could lead to timeouts or out-of-memory errors when compiling with the flag.
- Number of comments this week: 2
-
[TPU] Cannot initialize TPU system on TPUv4-Pod with TensorFlow 2.19: This issue involves a bug where the TPU system cannot be initialized on a TPUv4-Pod using TensorFlow 2.19, with different behaviors observed between two configurations:
tpu-vm-tf-2.19.0-pod-seandtpu-vm-tf-2.19.0-pod-pijt. Thetpu-vm-tf-2.19.0-pod-pijtconfiguration gets stuck at thetf.config.experimental_connect_to_cluster(cluster_resolver)step, while thetpu-vm-tf-2.19.0-pod-seconfiguration passes this step but encounters an error during thetf.tpu.experimental.initialize_tpu_system(cluster_resolver)step.- A user tagged several contributors to bring attention to another issue related to TPU initialization on TPUv4 Pod with TensorFlow 2.19, expressing gratitude with a series of rose emojis.
- Number of comments this week: 1
-
TF 2.19 GCC aarch64 Linux CUDA build failure: This issue involves a build failure when attempting to compile TensorFlow version 2.19.0 with CUDA support on an aarch64 Linux system using GCC 13.2.0, where the error does not occur for x86_64 CUDA or aarch64 CPU-only builds. The problem is characterized by a linking error related to the
libtensorflow_cc.so.2.19.0file, which is suspected to be associated with specific relocation issues in the MLIR components of TensorFlow.- The comment section references another issue, #78846, suggesting a potential connection or similar problem, but does not provide additional details or solutions.
- Number of comments this week: 1
-
Casting NaN to int32 yields different results on CPU and GPU: This issue highlights a bug in TensorFlow where casting a float32 tensor containing NaN values to int32 yields different results on CPU and GPU, with the CPU outputting -2147483648 and the GPU outputting 0. The inconsistency suggests a need for a warning or error when casting NaN to integer types, or for defining a consistent behavior across all devices.
- A user confirmed the issue by reproducing it on Colab using both TensorFlow 2.19.0 and the TF-nightly version, providing a gist link for reference.
- Number of comments this week: 1
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.
SystemErrorintf.ensure_shapeandtf.compat.v1.ensure_shapewhendtypeofshapeistf.uint64and its value is too large.: This issue involves 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 is reproducible in TensorFlow version 2.15 on a Linux Ubuntu 20.04 system, and it occurs when the specified shape value is excessively large, as demonstrated by the exampleshape = tf.constant([18446743219011059112, 1], dtype=tf.uint64).- Feature Request: Integrate different Digital Signal Processing into tf.signal: This issue is a feature request to integrate advanced Digital Signal Processing (DSP) functionalities into TensorFlow's
tf.signalmodule, similar to those available in the PyTorch ecosystem, particularly from the julius library. The request highlights the need for native tools within TensorFlow to perform complex audio data augmentation, which would enhance the capabilities for researchers and developers working on audio-related projects by streamlining workflows and reducing dependency on external libraries. - [DOCS] Missing complex input for Round op: This issue highlights a documentation bug in TensorFlow version 2.15.0, where the official documentation incorrectly states that the
Roundoperation can accept a complex tensor as input, but in practice, this results in an error unless the operation is applied separately to the real and imaginary parts. The user has reproduced this inconsistency using TensorFlow Nightly on a MacOS Sonoma platform with Python 3.9, and the error log indicates that theRoundoperation does not support complex data types on the available devices. - tf.raw_ops.Unbatch aborts with "Check failed: d < dims()": This issue pertains to 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 Python 3.11.8, and involves custom code that triggers the error when attempting to unbatch a tensor. Since there were fewer than 5 open issues, all of the open issues have been listed above.
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: 34
Summarized Issues:
- TensorFlow TPU Initialization Issues: The TPU system cannot be initialized on a TPUv4-Pod using TensorFlow 2.19, with configurations getting stuck or encountering errors during initialization steps. This affects the
tpu-vm-tf-2.19.0-pod-pijtandtpu-vm-tf-2.19.0-pod-seconfigurations, causing significant disruptions in TPU usage.
- TensorFlow Build and Compilation Failures: TensorFlow version 2.19.0 faces build failures with CUDA on aarch64 architecture using GCC 13.2.0, and xprof component compilation issues with GCC 14.2 on Ubuntu 25.04. These issues highlight compatibility problems with specific compiler versions and architectures.
- TensorFlow Segmentation Faults and Crashes: Multiple TensorFlow operations, including
tf.ragged.segment_ids_to_row_splits,tf.nn.conv2d_transpose, andtf.raw_ops.QuantizedConv2D, result in segmentation faults or crashes due to input shape issues or invalid operations. These bugs prevent proper error handling and disrupt model execution.
- TensorFlow Lite Inference and Model Conversion Issues: TensorFlow Lite faces inference crashes during high-volume operations and conversion failures with unsupported operations on edge devices. These issues hinder the deployment of models in resource-constrained environments and require detailed investigation and resolution.
- TensorFlow Quantization and Optimization Bugs: The TensorFlow model optimization library encounters critical bugs during the quantization of custom LSTM models, preventing the generation of optimized TFLite models. This impacts the deployment of efficient models in environments with limited resources.
- TensorFlow Operation Failures on CUDA: Operations like
tf.linalg.pinvandtf.linalg.svdcrash with SIGABRT errors due to out-of-memory conditions on CUDA, while they work on CPUs. These issues indicate potential memory management problems in TensorFlow's CUDA implementation.
- TensorFlow Import and Runtime Errors: Importing TensorFlow on Windows with Python 3.12.4 results in ImportError due to DLL initialization failures, and AttributeError occurs with Python Protobuf versions 6.30 or higher. These errors disrupt TensorFlow's usability on specific platforms and configurations.
- TensorFlow Documentation and Link Issues: A broken link in the TensorFlow pip installation guide leads to a "Page not found" error, indicating a documentation bug. This issue requires correction to ensure users can access necessary resources for installation.
- TensorFlow Operation Crashes with Invalid Inputs: Operations like
tf.raw_ops.Conv3DBackpropInputV2andtf.raw_ops.MaxPoolcrash due to malformed inputs or invalid data formats, highlighting the need for better input validation and error handling in TensorFlow.
- TensorFlow Floating Point and Arithmetic Exceptions: Enabling oneDNN optimizations in TensorFlow 2.20.0-dev20250516 causes floating point exceptions in operations like
AvgPool3DGradandMaxPool3D, indicating potential issues with arithmetic operations under specific conditions.
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: 25
Summarized Issues:
- Compilation and Build Issues: These issues involve problems encountered during the compilation and build process of TensorFlow on various platforms. One issue describes a bug when compiling TensorFlow 2.19.0 with MKL support on an ARM SVE platform using Clang, where the build fails due to a source code error. Another issue involves difficulties installing TensorFlow 2.18.1 on Centos 7 with Bazel, where users face challenges with protobuf version compatibility.
- GPU and CUDA Related Bugs: These issues highlight bugs related to GPU operations and CUDA compatibility in TensorFlow. One issue reports an illegal GPU memory access in TensorFlow 2.18.0 on Ubuntu 20.04 with CUDA 12.5.1, causing an out-of-bounds write error. Another issue involves a crash when using
tf.scatter_ndwith incorrect indices input in TensorFlow 2.19, leading to a CHECK-failure.
- Memory Leaks and Performance Issues: These issues describe memory leaks and performance problems in TensorFlow's image processing and data handling functions. One issue reports a memory leak in the image processing module when processing large image batches, especially with GPU acceleration. Another issue highlights a critical memory leak in the
tf.datapipeline when using themaptransformation with parallel calls, leading to out-of-memory errors.
- Mixed Precision and Apple M1 Crashes: These issues involve crashes when using mixed precision in custom training loops on Apple M1 hardware. The use of
tf.keras.Model.fit()with mixed precision causes the training process to crash, despite expectations for normal operation.
- Feature Requests and Compatibility Issues: These issues include feature requests and compatibility problems with TensorFlow on different platforms and Python versions. One request is for TensorFlow to support Arch Linux due to dependency issues with its rolling release model. Another request is for TensorFlow to support Python 3.13.x, as users encounter compatibility errors.
- Documentation and Usability Concerns: These issues address gaps in documentation and usability challenges in TensorFlow. One issue highlights a lack of clear guidelines for documenting new features, complicating contributors' and reviewers' tasks. Another issue reports a broken link in the Getting Started guide, leading to user frustration.
- Memory Management and Training Issues: These issues describe memory management problems during training sessions in TensorFlow. One issue reports unbounded memory usage during training sessions exceeding 50 epochs, leading to out-of-memory errors. Another issue involves a memory leak when using custom loss functions in the Keras API, unlike when using built-in loss functions.
- Platform-Specific Bugs: These issues involve platform-specific bugs in TensorFlow. One issue describes a bug where the
tf.raw_ops.Mfccoperation causes a segmentation fault instead of handling an error with a Python exception. Another issue involves an inability to create a Hexagon delegate for TensorFlow Lite on Android 10 due to file access errors.
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.
-
- Toxicity Score: 0.55 (Accusatory language, Defensive tone, Potential for escalation)
- This GitHub conversation involves a user expressing concern over repeated issues with an application, which is met with a terse response from another user accusing them of spamming. The tone is defensive and dismissive, with a potential for escalating tension due to the accusatory language used.
-
Bug: tf.keras.Model.fit() crashes with mixed precision in custom training loops on Apple M1
- Toxicity Score: 0.65 (Accusation of spamming, dismissive tone, defensive response.)
- This GitHub conversation involves a user expressing frustration over a recurring issue, with another user responding by accusing them of spamming. The tone is tense, with the initial comment being dismissive and the response being defensive. The trigger of tension appears to be the repeated mention of the issue without a perceived constructive contribution.
-
Memory leak when using custom loss functions in Keras API
- Toxicity Score: 0.65 (Dismissive response, accusation of spamming, invalidation of concern.)
- This GitHub conversation involves a user expressing frustration over a perceived issue with memory leaks when using custom loss functions in the Keras API, which is met with a dismissive response from another user who accuses them of spamming. The tone is tense, with the initial user's concern being invalidated, leading to potential conflict.
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. Eigen SVE build fix.: This pull request addresses a build issue in TensorFlow when using SVE-optimized Eigen by adding a neon header and implementing a new casting operation for converting BF16 to float using SVE, as evidenced by the error encountered during the compilation of sparse_matmul_op.cc and the subsequent commits that implement the BF16 to float SVE conversion and a clang fix.
- URL: pull/93811
- Merged: No
2. Tflite wheel armcortexa9 armv7 vfpv3: This pull request introduces a new armhf_vfpv3 build target to generate TensorFlow Lite wheels compatible with ARMv7-A architectures, such as Cortex-A9, with VFPv3 FPU support, addressing compatibility issues that caused illegal instruction errors on these devices due to the default armhf build targeting VFPv4.
- URL: pull/93941
- Merged: No
3. Stable delegate python api: This pull request introduces the capability to load TensorFlow Lite Stable Delegates from JSON configuration files via the Python Interpreter API by adding an experimental_stable_delegates argument to the Interpreter constructor, extending the C++ backend to parse and load delegates, exposing this functionality through pybind11, and updating Bazel build files to support delegate loading and JSON parsing.
- URL: pull/93850
- Merged: No
- Associated Commits: 796f2
Other Open Pull Requests
- Update aarch64-compatible oneDNN and ACL versions: This pull request updates the aarch64-compatible oneDNN to version 3.7 and the ACL to version 24.12. The update aims to enhance performance, improve memory management, and address numerous bugs from outdated versions, while also fixing unit test failures caused by a previous patch.
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: 8
Key Closed Pull Requests
1. Add complex64 and complex128 support to tf.truncatediv: This pull request adds support for complex64 and complex128 data types to the TensorFlow operation tf.math.truncatediv, addressing a previous runtime error by implementing a proper truncation strategy for complex division, registering necessary kernels, updating the Python API for consistency, and including unit tests to ensure correctness and alignment with the documentation.
- URL: pull/93125
- Merged: No
2. Update README.md - remove old links: This pull request involves updating the README.md file in the TensorFlow project by removing outdated links to the Continuous Integration (CI) system, as indicated by the commit messages and the fact that it was not merged.
- URL: pull/93460
- Merged: No
3. Readme cleanup: This pull request involves cleaning up the README file by removing outdated and non-functional links, as indicated by the commit message and the associated commit URL, although it was not merged into the main project.
- URL: pull/93466
- Merged: No
- Associated Commits: 8b7e5
Other Closed Pull Requests
- Updating and formatting the readme.md file: Two pull requests focused on updating and formatting the readme.md file in the TensorFlow project. Despite their efforts to improve documentation clarity, neither of these pull requests was merged into the main codebase.
- Facilitating the importing process: A pull request aimed to enhance the importing process by attempting to load PyInfo from the rules_python module. This change was proposed to streamline the import functionality, but it was not merged into the main project.
- New feature proposal: chatTensorFlow: A pull request proposed a new feature called chatTensorFlow, an Agentic RAG-based application for interacting with TensorFlow's documentation using natural language. This feature aimed to enhance user engagement with documentation, but it was not merged.
- Support for ARMv7-A architecture in TensorFlow Lite: A pull request added a new case to the
download_toolchains.shscript to support building TensorFlow Lite wheels for ARMv7-A architecture with VFPv3 FPU. This update addressed illegal instruction errors on Cortex-A9 by differentiating from the default armhf target.
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.
Based on our analysis, there are no instances of toxic discussions in the project's open or closed pull requests from the past week.
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 |
|---|---|---|---|---|
| mihaimaruseac | 0 | 0 | 0 | 27 |
| Venkat6871 | 1 | 0 | 0 | 22 |
| Redempt1onzzZZ | 1 | 0 | 9 | 12 |
| ShengDong207 | 0 | 0 | 19 | 0 |
| SilentTester73 | 0 | 0 | 12 | 1 |
| RahulSundarMCW | 10 | 2 | 0 | 0 |
| SwuduSusuwu | 0 | 0 | 2 | 8 |
| KAVYANSHTYAGI | 5 | 1 | 0 | 3 |
| MichaelHudgins | 5 | 3 | 0 | 0 |
| cx104906 | 0 | 0 | 6 | 1 |