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 compatibility. Key updates also include runtime support for the bfloat16 data type in the tfl.Cast operation, alongside the discontinuation of publishing libtensorflow packages, which 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.

1. Basic tf.function example fails with "Failed call to cuInit" on CPU only machine in Python interactive mode: This issue describes a bug where a basic TensorFlow function example fails with a CUDA initialization error on a CPU-only machine when run in Python interactive mode using TensorFlow 2.20 and Python 3.13. The user reports that the error occurs despite no GPU being present, and the same example works when executed from a script file rather than interactively.

   • The comments clarify that the issue does not occur with an older Python and TensorFlow version (3.10.19 and 2.17.0), and that the problem specifically arises only in Python interactive mode, suggesting that interactive mode may not be fully supported or stable for TensorFlow in these versions.
   • Number of comments this week: 3

2. Autograph could not transform function and running as-is: This issue reports a bug where the Autograph feature in TensorFlow fails to transform a function and instead runs it as-is, specifically encountered during work related to converting JAX models to TensorFlow SavedModel format. The user provides detailed environment information, code to reproduce the problem, and logs, seeking help to understand and resolve the issue.

   • The comments include a notification that changes have been made addressing the issue and a request for clarification on whether additional packages are needed to run the code properly, along with an inquiry about the root cause and potential performance implications.
   • Number of comments this week: 2

3. Update curl dependency to 8.16.0+ to address CVE-2025-0725 and CVE-2025-9086: This issue reports that TensorFlow currently depends on curl version 8.11.0, which contains two high-severity security vulnerabilities (CVE-2025-0725 and CVE-2025-9086) that are actively flagged by security scanners. The issue requests updating the curl dependency to version 8.16.0 or later to remediate these vulnerabilities and improve the security posture of TensorFlow deployments.

   • The single comment requests the submitter to create a pull request and tag the commenter for a quick review, indicating readiness to assist with the fix once a PR is submitted.
   • Number of comments this week: 1

4. Dataset.batch() crashes with core dump when using extremely large batch_size (INT64_MAX): This issue describes a bug in TensorFlow where using an extremely large batch size (specifically INT64_MAX) with the Dataset.batch() function causes a core dump due to an integer overflow during tensor shape calculations. The expected behavior is that the function should validate the batch size parameter, raise a clear and appropriate error for invalid values, and avoid crashing the entire process.

   • The comment confirms the crash occurs with very large batch sizes when drop_remainder=True, even in TensorFlow 2.20, and suggests computing the dataset’s cardinality to ensure the batch size is less than 2^63 − 1 to avoid the crash, providing a working example with a smaller batch size.
   • Number of comments this week: 1

5. [XLA/GPU] ConcatV2 op kernel constraint mismatch when using jit_compile=True with Python control flow: This issue describes a bug encountered when using TensorFlow's XLA compilation mode with jit_compile=True on a model that includes Python control flow operations like filter, map, and zip followed by a tf.concat operation. The error arises because the ConcatV2 GPU kernel has a type constraint mismatch, specifically expecting a different data type attribute than what is provided, causing the operation to fail during XLA compilation while it works fine in eager execution.

   • The single comment mentions recent changes made to the core code and requests a review and feedback from collaborators, indicating ongoing development and collaboration to address or investigate the issue further.
   • 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 the user is encountering a warning that TensorRT cannot be found when running TensorFlow on an Ubuntu 22.04 system with an NVIDIA RTX 3050 Ti GPU. Despite using the NVIDIA driver version 535 and CUDA 12.4, and attempting multiple reinstallations, the user is unable to resolve the TensorRT detection error, which is impacting their ability to proceed with machine learning development.
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 specifically in eager execution mode, as demonstrated by a shape tensor example with a large uint64 value, causing the built-in isinstance function to fail internally.
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 comprehensive audio data augmentation tools within TensorFlow compared to PyTorch and suggests that adding these capabilities would enhance audio model development by enabling native, efficient preprocessing and augmentation workflows.
4. [DOCS] Missing complex input for Round op: This issue highlights a documentation bug where the TensorFlow Round operation is described as supporting complex tensor inputs, but in practice, attempting to use a complex tensor with this operation results in an error. The user reports that they must manually apply the Round operation to the real and imaginary parts separately, indicating a discrepancy between the official documentation and the actual functionality.
5. tf.raw_ops.Unbatch aborts with "Check failed: d < dims()": This issue reports a bug in TensorFlow version 2.17 where the tf.raw_ops.Unbatch operation aborts with a "Check failed: d < dims()" error, causing the program to crash. The problem occurs when running a specific code snippet on Linux Ubuntu 20.04.3 with Python 3.11.8, and it has been reproduced using TensorFlow Nightly builds, indicating a persistent failure in handling certain tensor inputs.

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: 26

Summarized Issues:

• XLA JIT Compilation Errors and Limitations: Multiple issues report bugs and limitations when using @tf.function(jit_compile=True) with TensorFlow's XLA compiler. These include type constraint mismatches, symbolic tensor usage errors, dynamic shape incompatibilities, unsupported numpy() calls on variables, inaccessible tensor scope errors, and dictionary flattening type errors, all causing failures that do not occur in eager execution mode.
• [issues/105131, issues/105133, issues/105143, issues/105145, issues/105150, issues/105151, issues/105332, issues/105333]

• TensorFlow Build and Compilation Failures: Several issues describe build failures for TensorFlow Lite and TensorFlow on different platforms due to missing files or headers, such as missing Bazel BUILD files for iOS arm64, and missing protobuf headers on Ubuntu for Android builds. These prevent successful compilation and deployment on target platforms.
• [issues/105127, issues/105209]

• Integer Overflow and Large Value Handling Bugs: There are multiple bugs caused by improper handling of extremely large integer values (e.g., INT64_MAX, INT_MAX) in various TensorFlow operations. These include fatal assertion failures, segmentation faults, memory corruption, and crashes during operations like Dataset.batch(), tf.raw_ops.AsString, intra_op_parallelism_threads setting, TensorListSetItem, and SparseMatrixSparseCholesky.
• [issues/105098, issues/105292, issues/105297, issues/105302, issues/105303]

• TensorFlow Runtime and Kernel Errors with Python Control Flow and Custom Classes: Issues arise when combining Python builtins like filter, map, and zip with TensorFlow operations such as tf.concat, especially under XLA compilation, causing nondeterministic errors and kernel dtype mismatches. Additionally, using custom Python classes that create tf.constant or tf.Variable inside Keras models leads to NotImplementedError during XLA compilation due to unsupported numpy() calls.
• [issues/105131, issues/105150, issues/105151]

• Performance Issues in TensorFlow Components: The ImageProcessor component on Android devices exhibits slow preprocessing and postprocessing times for full-HD FP32 images, significantly impacting the overall runtime of models despite efficient inference.
• [issues/105217]

• TensorFlow Autograph and Model Conversion Failures: Autograph fails to transform functions properly during the conversion and saving of JAX-trained models using jax2tf, resulting in unoptimized execution of the function.
• [issues/105257]

• TensorFlow Lite Quantization Bugs: The QuantizationDebugger in TensorFlow Lite incorrectly quantizes transpose convolution weights to int8 while keeping inputs in float32, causing invalid quantization mismatches that should be resolved by keeping weights in float32.
• [issues/105281]

• TensorFlow MKL Backend Crashes: Enabling MKL in TensorFlow 2.21.0 causes fatal crashes in tf.nn.conv1d_transpose operations with dynamic batch sizes due to invalid dimension errors during shape inference in MKL convolution gradient input operations.
• [issues/105331]

• Requests for Diagnostic Information on TensorFlow Lite Issues: There are requests for detailed system information, reproducible code, and logs to diagnose problems related to TensorFlow Lite on Android devices and Google Play Services SDK.
• [issues/105313, issues/105314]

• Miscellaneous Discussions and Code Examples: Some issues contain brief comments or discussions unrelated to bugs, including a Python function example for uploading files to Google Cloud Storage and references to external pull request discussions.
• [issues/105298, issues/105340, issues/105341]

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: 17

Summarized Issues:

• CUDA kernel memory errors: Multiple issues report illegal out-of-bounds memory accesses and atomic operation errors in CUDA kernels during GPU execution, causing runtime errors detected by Compute-Sanitizer. These bugs affect various operations such as Bucketize, ClipByValue, DebugNumericSummaryV2, and MaxPoolGradWithArgmax on Ubuntu 22.04 with CUDA 12.5.1 and cuDNN 9.2.1, indicating widespread memory safety problems in GPU kernels.
• [issues/103469, issues/103470, issues/103994, issues/103998]

• CUDA kernel synchronization and compatibility issues: There is a risk of premature deallocation of input tensors in GPU kernels due to missing explicit CUDA stream synchronization, potentially leading to memory access errors or race conditions. Additionally, TensorFlow 2.20.0 experiences significant delays caused by just-in-time compilation of CUDA kernels on Kali Linux under WSL2 with NVIDIA RTX PRO Blackwell GPUs and CUDA 12.8, highlighting compatibility problems with newer CUDA versions and GPU architectures.
• [issues/103434, issues/103531]

• Build and linking problems on Linux and macOS: Several issues describe build failures and linking conflicts, including a missing LC_UUID load command error on macOS during Bazel builds, conflicts between system-installed gRPC and TensorFlow Lite's absl-cpp versions on Linux, and the absence of a package config file for TensorFlow Lite on Linux, all of which hinder proper compilation and integration. These problems suggest the need for improved dependency management and packaging in TensorFlow builds.
• [issues/103536, issues/103538, issues/103590]

• Documentation inaccuracies in LinearOperator error handling: The documentation for error conditions in tf.linalg.LinearOperatorScaledIdentity and tf.linalg.LinearOperatorCirculant functions requires correction and clarification to properly explain when ValueErrors are raised due to invalid parameter combinations. This lack of clarity can mislead users about the expected behavior and constraints of these linear operator classes.
• [issues/103911, issues/103915]

• Bazel build failures for TensorFlow Lite Android packages: Attempts to build the "tensorflow-lite-select-tf-ops.aar" package on Ubuntu 22.04 and similar environments fail due to ambiguous Bazel attribute matches and missing pthread functions, with only partial workarounds available. This issue complicates building TensorFlow Lite Android artifacts and requires further resolution.
• [issues/104763]

• TensorFlow runtime import and execution errors: Importing TensorFlow on Windows 10 with Python 3.11.9 fails due to a DLL load failure in the native runtime, preventing module initialization. Additionally, using symbolic tensors in Python control flow within Keras models causes OperatorNotAllowedInGraphError during XLA compilation with jit_compile=True, reflecting limitations in graph mode execution and runtime environment stability.
• [issues/105113, issues/105149]

• GPU multi-device training incompatibility: LSTM layers fail to run on multiple NVIDIA GPUs with TensorFlow 2.17.0 using CUDA 12.9 and cuDNN 9.*, causing training to abort due to incompatibilities in the CudnnRNN operation. This issue highlights version mismatches and operational failures in multi-GPU setups.
• [issues/105092]

• Miscellaneous bug reports and environment-specific issues: Some issues report unclear or environment-specific bugs, including a Docker Hub related problem reproducible with TensorFlow Nightly on Ubuntu 22.04, and a nonsensical spam issue with incomplete details submitted to the project. These reports indicate ongoing challenges in environment compatibility and issue triage.
• [issues/105065, issues/104982]

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. Fix/threadpool intmax segfault #105297 : This pull request fixes a segmentation fault in TensorFlow caused by setting intra_op_parallelism_threads to INT_MAX by adding an upper-bound validation in the ThreadPool constructor that clamps the number of threads to a maximum of 256, preventing excessive thread creation, memory exhaustion, and crashes while logging a warning when the limit is enforced.

• URL: pull/105310

• Merged: No

• Associated Commits: adaef, 055b6, be02d, b14b0

2. CUDAfix: This pull request addresses a critical bug in TensorFlow's tf.raw_ops.GatherV2 operation where invalid axis parameters caused CUDA device context initialization failures by adding proper axis bounds checking after axis normalization to prevent out-of-bounds access, thereby producing clear InvalidArgumentError messages before any CUDA operations and maintaining backward compatibility, along with comprehensive tests to cover various axis validation scenarios.

• URL: pull/105285

• Merged: No

• Associated Commits: 34c06, 388e6, bc503

3. Fix/concat nondeterministic: This pull request adds comprehensive regression tests and fixes for nondeterministic behavior and dtype coercion errors in tf.concat caused by mixing Python built-ins like filter, map, and zip with TensorFlow tensors, including demonstration tests, minimal tuple-flattening fixes, robust pure TensorFlow implementations, edge case handling, and verification of stability under XLA/JIT compilation.

• URL: pull/105248

• Merged: No

• Associated Commits: 03045, 934f3

Other Open Pull Requests

• XLA compilation and symbolic tensor conditionals: This set of pull requests improves XLA compatibility by addressing errors caused by Python conditionals with symbolic tensors and fixing shape-inference inconsistencies that lead to premature static specialization of dynamic dimensions. They provide regression tests, alternative usage patterns, and ensure dynamic shapes are preserved during JIT compilation to prevent runtime errors.
• pull/105251, pull/105250

• Segmentation fault and bounds checking fixes: Multiple pull requests add comprehensive bounds checking to prevent segmentation faults in TensorListSetItem and AsString operations by validating index ranges and width parameters, thereby avoiding integer overflow, excessive memory allocation, and crashes. These changes include new unit tests and error handling to improve stability and robustness.
• pull/105318, pull/105345

• FastImageProcessor enhancements: Two pull requests introduce and optimize the FastImageProcessor for Android and core TensorFlow Lite Java, significantly improving FP32 image preprocessing and postprocessing performance by reducing allocations and buffer copies. They add support for dynamic image dimensions, fused operations, buffer reuse, and cross-platform compatibility with thorough testing and build integration.
• pull/105222, pull/105272

• XLA operation and kernel fixes: These pull requests fix kernel constraint mismatches in the XLA ConcatV2 operation by adding support for all float8 types and resolve XLA tracing errors by replacing eager-only calls with compatible tensor conversions, ensuring correct operation under JIT compilation.
• pull/105282, pull/105233

• Command-line and flag parsing improvements: One pull request improves the parsing logic of the --tf_xla_cpu_global_jit flag in the Grappler remapper optimizer to handle different flag separators, while another introduces a new command-line tool for inspecting TensorFlow and Keras model architectures without Python code, supporting multiple formats and export options.
• pull/105000, pull/105180

• Code quality and dependency updates: A pull request fixes typos, corrects constant names, and improves error message formatting across multiple files to enhance code quality and consistency. Another updates the curl dependency to a newer version to fix two high severity security vulnerabilities related to gzip decompression and cookie path handling.
• pull/105204, pull/105118

• Tosa conversion improvements: This pull request enhances the TosaConvertTFLUInt8 pass by correctly handling quantized unsigned 8-bit values and fixing function signature updates to prevent illegal IR errors during tensor reshaping.
• pull/105178

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: 9

Key Closed Pull Requests

1. CVE-2023-37920 @ requirements_mac.txt: This pull request updates the certifi dependency to version 2023.7.22 or newer in the Mac build environment to remove the untrusted e-Tugra root certificate from the root store, thereby mitigating the risk of Man-in-the-Middle (MITM) attacks through certificate spoofing as addressed in CVE-2023-37920.

• URL: pull/104892

• Merged: 2025-11-24T17:35:44Z

• Associated Commits: 813bf, de857

2. Dev: This pull request titled "Dev" includes commits that add a testing command and a custom YAML file to the TensorFlow project but was not merged.

• URL: pull/105246

• Merged: No

• Associated Commits: a65eb, 49f84

3. Fix the doc of error raised in func tf.linalg.LinearOperatorScaledIdentity: This pull request fixes the documentation regarding the error raised in the function tf.linalg.LinearOperatorScaledIdentity in the TensorFlow project.

• URL: pull/103912

• Merged: 2025-11-24T18:13:34Z

• Associated Commits: 9c654

Other Closed Pull Requests

• Documentation fixes: This topic includes pull requests that improve or add documentation content to enhance clarity and completeness. One pull request fixes the docstring for the error raised in tf.linalg.LinearOperatorCirculant, while another adds introductory content to the previously empty introduction.md file in the "Get Started" docs.
• [pull/103916, pull/104945]

• Code cleanup: This topic covers pull requests aimed at minor code cleanups to improve code quality without changing functionality. One pull request removes an extra semicolon in tfe_wrapper_monitoring_reader.cc to clean up the code.
• [pull/104548]

• Build and configuration updates: This topic groups pull requests that update build files and CI configurations to support new architectures and improve infrastructure robustness. One pull request updates workspace0.bzl to integrate llvm21 support for aarch64, and another enhances the BUILD configuration for the requirements updater by adding visibility settings, CI tags, and extended timeouts.
• [pull/104910, pull/104931]

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.