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 and the discontinuation of standalone libtensorflow package publishing, while still allowing unpacking from 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. [TYPE:BUILD/INSTALL] [SUBTYPE:WINDOWS] Failed to load the native TensorFlow runtime.: This issue concerns a failure to load the native TensorFlow runtime on a Windows system, resulting in an ImportError related to a DLL initialization failure when importing a TensorFlow internal module. The user reports that the MSVC 2019 redistributable is installed and the system architecture is 64-bit, but the error persists during the import of TensorFlow, affecting a student's attempt to install a Python library.

   • The comments focus on gathering more information about the TensorFlow version, system environment, and installation steps, with suggestions about common causes such as missing redistributables, CPU instruction support, and architecture mismatches; the user clarifies the TensorFlow version and system details, noting the Windows edition correction from Educational to Pro.
   • Number of comments this week: 3

2. [TYPE:BUG] [Bug] tf.where returns None gradient for differentiable expression: This issue reports that using tf.where with differentiable expressions results in a None gradient, despite the underlying function being mathematically differentiable, causing silent failure in gradient computation. The problem arises because tf.where evaluates both branches and does not create a piecewise-differentiable function, leading TensorFlow's automatic differentiation to return None when the condition depends on the variable being differentiated.

   • The comments clarify that this behavior is expected due to how tf.where operates as an element-wise selector rather than true control flow, and recommend using arithmetic masking or tf.cond to preserve gradient flow; they also suggest defining custom gradients as an alternative and note that a warning from TensorFlow would be helpful to avoid confusion.
   • Number of comments this week: 2

3. [TYPE:BUG] [COMP:OPS] [2.20.0] tf.math.cumulative_logsumexp produces NaN for inputs containing +inf: This issue reports that the TensorFlow function tf.math.cumulative_logsumexp produces NaN values when the input contains multiple +inf values, despite the mathematically expected output being +inf. The problem arises from the internal computation of pairwise log-add-exp, where subtracting inf - inf leads to NaN, and a proposed fix involves adding a special-case guard to handle inputs with infinite values correctly.

   • The comment explains the root cause as the subtraction of infinite values in the log-sum-exp identity and provides a code snippet for a safe implementation that handles +inf inputs properly, ensuring numerical stability and correct results consistent with other libraries like JAX and NumPy.
   • Number of comments this week: 1

4. [TYPE:BUG] [COMP:OPS] [COMP:XLA] [2.21.0] tf.gather returns a value for an out-of-bounds index under tf.function(jit_compile=True) while eager rejects it: This issue reports a bug where the tf.gather function behaves inconsistently between eager execution and compiled execution with jit_compile=True: it raises an error for out-of-bounds indices in eager mode but returns a value instead when compiled. The user demonstrates that under compilation, an out-of-bounds index returns an unexpected value rather than raising an error as documented for CPU devices.

   • A commenter confirmed the issue by reproducing it on Colab with TensorFlow versions 2.21.0 and nightly, providing a shared notebook for reference.
   • Number of comments this week: 1

Since there were fewer than 5 open issues, all of the open issues have been listed above.

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.

As of our latest update, there are no stale issues for the project this week.

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: 7

Summarized Issues:

• Quantization and Computation Fallback: The TFLite int16x8 conversion process incorrectly inserts Quantize and Dequantize operations around the SQRT function, causing an unexpected fallback to float computation instead of keeping the operation within the quantized graph. This issue affects the efficiency and correctness of quantized model execution.
• issues/115046

• Memory Safety and Overflow Risks: The TfLiteTensorResizeMaybeCopy function has a potential integer overflow when adding 16 to num_bytes without proper checks, which can cause allocation size wraparound and result in an undersized heap buffer. This leads to unsafe tensor size accounting and risks memory corruption or crashes.
• issues/115068

• Numerical Stability and NaN Propagation: The tf.math.cumulative_logsumexp function produces NaN values when the input contains multiple +inf values due to an internal computation of inf - inf, which should mathematically yield +inf but currently results in NaN propagation. This bug affects the reliability of computations involving infinite values.
• issues/115091

• Gradient Computation Failures: Using tf.where with differentiable expressions in TensorFlow 2.19.0 results in a None gradient because both branches are evaluated and gradient backpropagation is mishandled. This causes silent failures in gradient computation despite the function being mathematically differentiable.
• issues/115116

• Indexing and Error Handling Inconsistencies: The tf.gather function incorrectly returns a value for an out-of-bounds index when used inside a tf.function with jit_compile=True, whereas in eager mode it correctly raises an error. This inconsistency can lead to unexpected behavior in compiled TensorFlow functions.
• issues/115254

• Documentation and Tutorial Clarity: The TensorFlow "Tensor" tutorial has documentation issues including a missing NumPy import statement that causes errors for beginners, and unclear explanation in the "Single-axis indexing" subsection where "Every other item" should be replaced with "Every odd item" to accurately describe slicing behavior. These issues reduce the tutorial's clarity and usability.
• issues/115277, issues/115278

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: 1

Summarized Issues:

• Memory Leak in tf.keras.Model Training Loop: This issue describes a progressive memory leak that occurs when repeatedly creating, training, and deleting tf.keras.Model instances inside a loop. Despite attempts to clear sessions, delete models, and force garbage collection, memory usage steadily increases across different operating systems, Python versions, and TensorFlow variants.
• issues/113529

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: 11

Key Open Pull Requests

1. Fix tf.ragged.constant shape inference for uniform outer dimensions #40159: This pull request fixes the shape inference in tf.ragged.constant for inputs with uniform outer dimensions by adding uniform dimension detection, stabilizing ragged rank calculation, handling deeply nested empty structures to prevent crashes, and ensuring consistent type conversion for mixed np.ndarray and list inputs.

• URL: pull/115237

• Associated Commits: 06a10, f7217, c90a5

2. Docs/clarify where gradienttape unconnected: This pull request clarifies the documentation around issues with unconnected targets in TensorFlow's GradientTape, particularly when building targets outside the tape causing None returns, and suggests using tf.cond, masking, or custom gradients for piecewise intent as helpful solutions.

• URL: pull/115258

• Associated Commits: b904d, 725cf, 87baa

3. Fix integer overflow in stablehlo_reduce_window dimension computation: This pull request fixes an integer overflow in the stablehlo_reduce_window dimension computation by adding overflow detection and range validation to prevent silent truncation of large dimension values, thereby avoiding heap buffer overflows and ensuring that tensor allocation fails gracefully with an error instead of corrupting memory.

• URL: pull/114990

• Associated Commits: f5d43

Other Open Pull Requests

• Integer overflow and memory safety in TFLite: Multiple pull requests address integer overflow and memory safety issues in TFLite by adding comprehensive overflow checks in output dimension computations and safeguarding against allocation size overflow during tensor resizing. These changes prevent silent truncation errors and buffer overflows, enhancing the robustness of TFLite operations.
  • pull/115031, pull/115069

• Bug fixes in linear algebra and cumulative operations: Pull requests fix bugs in tf.linalg.eigh_tridiagonal that caused crashes with certain select parameters and dynamic input shapes, and also correct the tf.math.cumulative_logsumexp function to handle multiple +inf inputs properly by short-circuiting to +inf. These fixes improve stability and correctness in linear algebra and cumulative log-sum-exp computations.
  • pull/115086, pull/115123

• Improved error handling in AutoGraph: One pull request enhances AutoGraph's error handling for conditional expressions with mismatched branches by raising a user-facing ValueError instead of an internal assertion error, and adds a regression test for this specific pattern. This change improves the clarity and usability of error messages in AutoGraph.
  • pull/115111

• Fixes for XLA compilation and operator quantization: Pull requests fix an XLA compilation failure in tf.random.shuffle with jit_compile=True by correcting the sort comparison type, and fix int16x8 quantization for the SQRT operator in TFLite by properly registering the operator to allow quantization. These fixes ensure correct compilation and quantization behavior in XLA and TFLite.
  • pull/115189, pull/115190

• XLA GatherV2 operation bounds checking: A pull request adds bounds checking to the XLA GatherV2 operation by clamping out-of-bounds indices, creating a validity mask, and zeroing invalid results to prevent silent data corruption. This ensures consistency with TensorFlow's eager mode and improves operation safety.
  • pull/115255

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: 1

Key Closed Pull Requests

1. Fix int16x8 quantization for SQRT op in TFLite: This pull request fixes the int16x8 quantization for the SQRT operator in TFLite by properly registering BuiltinOperator_SQRT in the operator property switch to prevent it from being incorrectly marked as non-quantizable, thereby avoiding unnecessary Dequantize/Quantize operations and enabling the use of the existing int16 quantized kernel.

• URL: pull/115135

• Associated Commits: 1477b, 02587, 3314c, 5a047, 8967c

• Associated Commits: 1477b, 02587, 3314c, 5a047, 8967c

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.

Access Last Week's Newsletter:

• Link