Weekly GitHub Report for Tensorflow: October 06, 2025 - October 13, 2025 (12:06:47)

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
  • 1.2. Other Noteworthy Updates
• II. Issues
  • 2.1. Top 5 Active Issues
  • 2.2. Top 5 Stale Issues
  • 2.3. Open Issues
  • 2.4. Closed Issues
  • 2.5. Issue Discussion Insights

• III. Pull Requests
  • 3.1. Open Pull Requests
  • 3.2. Closed Pull Requests
  • 3.3. Pull Request Discussion Insights

• IV. Contributors
  • 4.1. Contributors

I. News

1.1 Recent Version Releases:

The current version of this repository is v2.19.0

1.2 Version Information:

Released on March 5, 2025, TensorFlow version 2.19.0 introduces breaking changes to the tf.lite API, including the deprecation of tf.lite.Interpreter in favor of ai_edge_litert.interpreter and changes to certain C++ constants for improved API flexibility. Key updates also include runtime support for the bfloat16 data type in the tfl.Cast operation, alongside the discontinuation of standalone libtensorflow package publishing, with the package still 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.

1. [Bug] CUDA_ERROR_INVALID_PTX on new RTX 5080 GPU with Driver 581.29: This issue reports a bug where TensorFlow operations requiring JIT compilation fail on a new NVIDIA RTX 5080 Laptop GPU with the latest drivers, producing CUDA_ERROR_INVALID_PTX and CUDA_ERROR_INVALID_HANDLE errors. The problem occurs because the current TensorFlow pip binaries lack precompiled CUDA kernels for the NVIDIA Blackwell architecture (Compute Capability 12.0), causing the CUDA 13.x driver to reject the JIT-compiled PTX code.

   • The comment explains that the root cause is the absence of precompiled CUDA kernels for the new GPU architecture in TensorFlow’s pip packages, leading to fallback JIT compilation that the CUDA 13.x driver does not accept, and directs the user to a related ongoing tracking issue for further updates and discussions.
   • Number of comments this week: 1

2. SIGSEGV on armv7a devices: This issue reports segmentation faults occurring on armv7a devices when running TensorFlow Lite 2.16.2 compiled from source, specifically related to the StridedSlice operation in the native C++ code. The user is seeking assistance to identify the root cause of these crashes and to find a potential solution, providing a detailed backtrace of the failure.

   • The comment identifies the crash as a native memory access error likely caused by an integer overflow or miscalculation in the StridedSlice kernel on 32-bit armv7a devices and requests a minimal reproducible example including the .tflite model file and input tensors to facilitate debugging and resolution.
   • Number of comments this week: 1

3. Error on load tensorflow: This issue reports a bug where TensorFlow fails to load its native runtime on a Windows 11 system using Python 3.13.7 and TensorFlow version 2.20, resulting in a DLL load failure error. The user requests assistance to resolve this compatibility or environment-related problem that prevents TensorFlow from initializing properly.

   • The comment requests additional environment details and installation steps, suggests possible causes such as missing MSVC redistributables, CPU instruction set incompatibility, or 32-bit Python/CPU issues, and points out that the issue may be a duplicate of previously reported problems.
   • Number of comments this week: 1

4. Fix the descriptions of limitation in several RFFT funcs: This issue addresses inaccuracies in the documentation of several real FFT (RFFT) functions in TensorFlow, specifically noting that the current descriptions do not mention the dtype limitations which cause errors when using unsupported complex types. The reporter provides multiple reproducible examples demonstrating the errors and requests that these dtype constraints be clearly documented to improve clarity for users.

   • A pull request has been opened to fix the documentation limitations as described, focusing solely on the dtype restriction issue; other related concerns mentioned in the issue remain unaddressed for now, with an invitation for further feedback if additional changes are needed.
   • Number of comments this week: 1

5. Incorrect result when using XLA (jit_compile=True) with tf.math.reduce_mean on int32 tensors: This issue reports a bug where using XLA compilation (jit_compile=True) with tf.math.reduce_mean on int32 tensors produces incorrect results due to integer overflow during the summation step. The user demonstrates that the mean computed with XLA differs significantly from the non-XLA version because the intermediate sum exceeds the int32 range, causing overflow and inconsistent results.

   • The comment explains that the sum of the input tensor exceeds the int32 limit, causing overflow in both XLA and non-XLA executions; however, XLA's optimization changes the addition order, leading to different overflow behavior and thus different results. The recommended fix is to use a wider accumulator type, such as int64, for intermediate sums to avoid overflow and ensure consistent results.
   • 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.

1. TF-TRT Warning: Could not find TensorRT: This issue describes a problem where TensorFlow on Ubuntu 22.04 cannot detect TensorRT despite having compatible NVIDIA drivers and CUDA versions installed, specifically with an RTX 3050 Ti GPU and CUDA 12.4. The user reports difficulties with driver versions, having to downgrade from the default installed NVIDIA 550 driver to version 535, and continues to receive a warning that TensorRT could not be found, impacting their ability to use TensorFlow efficiently.
2. SystemError in tf.ensure_shape and tf.compat.v1.ensure_shape when dtype of shape is tf.uint64 and its value is too large.: This issue reports a bug in TensorFlow where calling tf.ensure_shape or tf.compat.v1.ensure_shape with a shape tensor of type tf.uint64 containing very large values close to 2^64 triggers a SystemError and OverflowError. The problem occurs in eager execution mode, causing these APIs to fail when handling such large shape values, which is unexpected behavior and likely a defect in the library.
3. Feature Request: Integrate different Digital Signal Processing into tf.signal: This issue is a feature request proposing the integration of advanced digital signal processing (DSP) functionalities, similar to those found in the julius library, into TensorFlow's tf.signal module. The requester highlights the current lack of sophisticated audio data augmentation tools within TensorFlow compared to PyTorch and suggests that adding these capabilities would enhance audio model training by enabling native, efficient preprocessing and augmentation workflows.
4. [DOCS] Missing complex input for Round op: This issue reports a documentation bug concerning the Round operation in TensorFlow, where the official documentation states that complex tensors are supported as input, but in practice, attempting to use a complex tensor with the Round op results in an error. The user found that they must manually apply the Round operation separately to the real and imaginary parts of the complex tensor to achieve the expected behavior, indicating a discrepancy between the documented capabilities and the actual implementation.
5. tf.raw_ops.Unbatch aborts with "Check failed: d < dims()": This issue reports a bug in TensorFlow version 2.17 where the operation tf.raw_ops.Unbatch aborts with a fatal error message "Check failed: d < dims()" when executed with certain inputs. The problem occurs on Linux Ubuntu 20.04.3 LTS using Python 3.11.8, and the user has provided a minimal reproducible example demonstrating that the error causes the program to abort due to a tensor shape check failure.

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: 18

Summarized Issues:

• GPU and CUDA Compatibility Issues: Several issues report problems related to GPU support and CUDA compatibility, including errors from missing precompiled CUDA kernels for new GPU architectures causing kernel launch failures, and silent crashes on specific NVIDIA GPUs likely due to driver or cuDNN incompatibilities. These problems highlight challenges in ensuring TensorFlow's compatibility with the latest hardware and driver versions.
• issues/101746, issues/102065

• TensorFlow Lite Conversion and Usage Problems: Multiple issues involve difficulties with TensorFlow Lite, including user requests for help with model conversion failures, and several template issues lacking sufficient details to diagnose problems with the tflite_convert tool. These indicate ongoing challenges users face in converting and deploying models with TensorFlow Lite.
• issues/102051, issues/102052, issues/102053, issues/102054

• Numerical Accuracy and Overflow Bugs: Several bugs concern numerical inaccuracies and overflow issues, such as incorrect results from XLA-compiled reductions on int32 tensors due to overflow, overflow producing infinity in distributed float16 summations, and inconsistent inverse FFT results between high-level and low-level APIs. These issues reveal problems in numerical stability and precision in TensorFlow operations.
• issues/102043, issues/102046, issues/102047

• Data Type and Operation Inconsistencies: Some issues report inconsistent behavior in TensorFlow operations involving data types, including incorrect comparison results between scalars and RaggedTensors, and documentation lacking clarity on dtype limitations for RFFT functions causing runtime errors. These highlight the need for better dtype handling and documentation.
• issues/102027, issues/102044

• Image Processing Precision Loss: There is a reported issue where TensorFlow's RGB to HSV and back conversions are not numerically reversible for very large float32 values, resulting in significant precision loss. This points to limitations in the image processing functions when handling extreme input ranges.
• issues/102048

• Segmentation Fault on ARM Devices: One issue describes a segmentation fault occurring on armv7a devices when running TensorFlow Lite compiled from source, likely due to a memory access error in the C++ implementation of the StridedSlice operation. This indicates platform-specific stability problems in TensorFlow Lite.
• issues/101889

• DLL Load Failures on Windows: An issue reports failure to load TensorFlow's native runtime on Windows 11 with Python 3.13.7 due to a DLL load error related to the _pywrap_tensorflow_internal module, preventing proper import and usage. This reflects challenges in maintaining Windows compatibility with new Python versions.
• issues/101960

• LiteRT Runtime Support Clarification: There is a request for clarification on whether TensorFlow 2.20.0 includes internal support for the new LiteRT runtime intended to replace the deprecated tf.lite module, or if this integration is planned for a future release. This shows user interest in the roadmap for TensorFlow Lite's evolution.
• issues/102071

• Memory Page Alignment for Android Apps: One issue reports that the LiteRT Support 1.4.0 package ships a native library with 4 KB memory page alignment, which fails Google Play's new 16 KB alignment requirement, causing validation errors for apps targeting Android 15 or later. This highlights compliance issues with platform requirements.
• issues/101991

• Numerical Stability Differences with XLA: An issue describes a bug where tf.linalg.trace returns opposite infinity signs depending on whether XLA compilation is enabled, due to differences in numerical stability and summation order between eager and XLA backends. This points to subtle numerical behavior changes introduced by XLA.
• issues/102084

• Insufficient Issue Details Hindering Diagnosis: Several issues report bugs or problems but lack critical information such as system details, error outputs, reproducible code, or logs, making it difficult to diagnose and resolve the reported problems. This underscores the importance of detailed issue reporting for effective troubleshooting.
• issues/102055

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: 0

Summarized Issues:

As of our latest update, there were no issues closed in the project this week.

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: 18

Key Open Pull Requests

1. feat(keras): Add SeparableConv2DTranspose layer: This pull request introduces a new SeparableConv2DTranspose layer in Keras, including its operation registration, a reference CPU kernel, and the corresponding Python API, to provide an efficient and lightweight upsampling option particularly beneficial for generative models and autoencoders on mobile and edge devices.

• URL: pull/102104

• Merged: No

• Associated Commits: de50f, 663a1, f463d, ee36e, 9041e, 3d698, 59bed, 6ced3, 6dd4c

2. Fix: GPU Numerical Precision in TFLite Add+Mul Ops (Issue #66740): This pull request fixes a critical numerical precision bug in TensorFlow Lite's GPUv2 delegate by adding explicit convert_float4() calls in the OpenCL kernels for Add and Mul elementwise operations to ensure consistent 32-bit floating-point accuracy, resulting in a 50,000x reduction in max absolute error and full test pass rate with minimal performance impact.

• URL: pull/102117

• Merged: No

• Associated Commits: ac177, 097d0, 26f95, 9986a, 12cd6, 3d9ef, 0a728

3. Fix the descriptions of limitation in several RFFT funcs: This pull request aims to fix the descriptions of limitations in several RFFT functions by updating their API definition files to correct the docstrings, addressing issue #102027 without modifying other unrelated issues.

• URL: pull/102028

• Merged: No

• Associated Commits: 79e61, 255c5, 9a92b, 68698

Other Open Pull Requests

• Bug fixes in tensor operations and numerical stability: Multiple pull requests address bugs related to tensor operations and numerical stability, including fixing tf.nn.weighted_moments() to handle tensor-type axes correctly, resolving integer overflow in XLA-compiled reduce_mean by upcasting accumulators, and improving numerical precision in tf.image.rgb_to_hsv by using double precision. Additional fixes include preventing premature overflow in tf.distribute.Strategy.reduce("SUM") for float16 tensors and correcting integer truncation in tf.math.not_equal comparisons involving tf.RaggedTensor.
• pull/102086, pull/102095, pull/102098, pull/102099, pull/102096

• Axis parameter and validation improvements: Pull requests improve handling and validation of axis parameters in TensorFlow functions, including clarifying the axis attribute in tf.raw_ops.Dequantize() and correcting validation logic between dilation rate and strides in nn_impl.py.
• pull/102034, pull/102035

• Typographical and documentation corrections: Several pull requests fix typographical errors and improve documentation clarity, such as correcting confusing phrases in dense_attention.py, fixing misspellings like "Seperate" to "Separate," and enhancing the GitHub issue template formatting.
• pull/101806, pull/101905

• Runtime checks and validation in TensorFlow Lite kernels: A pull request introduces runtime checks in TensorFlow Lite's Transpose and ResizeBilinear kernels to ensure int8-quantized tensors have matching scale and zero_point parameters, aborting execution with an error if mismatches occur, and adds unit tests to verify this behavior.
• pull/101974

• Security and build configuration fixes: One pull request fixes a command injection vulnerability in the gRPC TPU worker service by replacing unsafe shell calls with safer alternatives, while another addresses a Google Play validation error on Android 15+ by modifying the core JNI library build configuration to set a 16KB memory page alignment.
• pull/102126, pull/102094

• Kernel and platform-specific bug fixes: A pull request adds a validation check to prevent zero-sized spatial dimensions in Keras MaxPool3D layers to avoid cuDNN crashes, and another fixes incorrect results for int64 tensors on Windows by patching minimum and maximum functors to handle 64-bit integer comparisons correctly under MSVC.
• pull/102100, pull/102101

• CMake build improvements for TensorFlow Lite dependencies: A pull request resolves a CMake build issue by adding install and export commands for critical TensorFlow Lite dependencies like XNNPACK and pthreadpool, ensuring proper installation and export for use in other CMake projects.
• pull/102102

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: 3

Key Closed Pull Requests

1. greptile-test: This pull request titled "greptile-test" appears to be a test-related update or experiment in the TensorFlow project, but it was not merged.

• URL: pull/101854

• Merged: No

• Associated Commits: ee0f2

2. TFLite: require matching int8 quantization params (scale & zero_point) for Transpose and ResizeBilinear: This pull request introduces runtime checks in TensorFlow Lite's Transpose and ResizeBilinear kernels to ensure that int8-quantized input and output tensors have matching scale and zero_point parameters, aborting execution with an error if mismatches are detected to prevent silent inaccuracies, and includes new unit tests verifying this behavior.

• URL: pull/101973

• Merged: No

• Associated Commits: a4142

3. Fix: GPU Numerical Precision in TFLite Add+Mul Ops (Issue #66740): This pull request addresses issue #66740 by fixing numerical precision errors in TensorFlow Lite's GPU delegate for Add and Mul operations through explicit conversion to float32 in OpenCL kernels, resulting in a 50,000x reduction in max absolute error and full test pass rates with minimal performance impact.

• URL: pull/102116

• Merged: No

• Associated Commits:

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
ILCSFNO	14	6	8	7
CodersAcademy006	9	9	0	9
khteh	0	0	7	3
kshiteej-mali	7	2	0	0
mihaimaruseac	0	0	0	9
kadirb4rut	3	2	0	1
LiSsHhUuAaIi	0	0	6	0
syan427	0	0	6	0
KamranBaig1122	2	2	0	1
Ma-gi-cian	2	1	0	2