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 a breaking change in the tf.lite API, with the C++ constants now as const references and a deprecation warning for the Python tf.lite.Interpreter redirecting users to ai_edge_litert.interpreter ahead of its removal in version 2.20. Additionally, the tfl.Cast operation now supports bfloat16 in the runtime kernel, and the libtensorflow packages have been discontinued from publishing but 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. [TYPE:BUG] [TF 2.19] [Bug] Logical flaw in auto_control_deps_utils.py: Resource categorization fails for unsorted _read_only_resource_inputs: This issue reports a bug in TensorFlow's auto_control_deps_utils.py where the functions responsible for categorizing resource inputs fail to correctly identify all read-only resources if the _read_only_resource_inputs attribute is unsorted, leading to false write dependencies and unnecessary serialization of operations. The root cause is a pointer-based linear scan that assumes sorted indices, causing misclassification when the attribute list is unsorted, which significantly impacts performance in complex models by injecting redundant control edges.

   • The comments confirm the root cause and propose fixes involving sorting or using a set for O(1) lookups to ensure correctness; a pull request was opened to address the bug. Further discussion explores a long-term architectural improvement using bitmask-based resource categorization for efficiency, but the consensus is to keep the current fix focused on correctness, deferring larger optimizations to future work.
   • Number of comments this week: 7

2. [TYPE:BUG] [COMP:OPS] Shape mismatch error message in tf.matmul is unclear for 2D x 1D input: This issue addresses the unclear error message produced by tf.matmul when multiplying a 2D matrix by a 1D vector with incompatible dimensions, which does not specify which dimensions are causing the mismatch. The user requests an improvement to the error message to explicitly indicate the inner dimension incompatibility, aiding developers in quickly identifying and resolving shape mismatches.

   • The comments explain the cause of the error and suggest reshaping the 1D vector to a 2D matrix to fix the issue. Several contributors emphasize the importance of clear error messages for debugging and compliance, especially in production environments, and express interest in improving the error message clarity, with some offering to collaborate on the fix.
   • Number of comments this week: 6

3. [TYPE:BUG] Missing depthwise conv 16x8 xtensa kernel: This issue addresses the lack of an optimized depthwise convolution kernel for the Xtensa platform when using 16-bit activations and 8-bit weights, causing the system to default to a less efficient reference kernel instead. The user highlights the need for leveraging the existing optimized implementation to improve performance on Xtensa devices.

   • The comments indicate that the issue was misplaced and has been transferred to a more appropriate repository, with one user confirming resolution and another discussing compliance requirements related to kernel specialization under the EU AI Act, ultimately suggesting the issue can be closed.
   • Number of comments this week: 4

4. [COMP:LITE] vryp: This issue is about a user reporting a problem related to TensorFlow Lite conversion but has not provided sufficient system information, TensorFlow version details, or reproducible code to help diagnose the problem. The issue template was left incomplete, making it difficult for maintainers to analyze and address the reported problem effectively.

   • The comments request the user to fill out the issue template with necessary details such as TensorFlow version and reproducible code to facilitate troubleshooting, and also suggest considering risk and regulatory impact labels for high-risk AI deployment scenarios.
   • Number of comments this week: 2

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: 8

Summarized Issues:

• Insufficient diagnostic information for TensorFlow Lite converter issues: A user reported problems with the TensorFlow Lite converter but did not provide enough system information, reproducible code, or logs, making it difficult to diagnose the issue. This lack of detail hinders effective troubleshooting and resolution of the problem.
• issues/110504

• Unclear error messages and dimension mismatch in matrix operations: The error message from tf.matmul when multiplying a 2D matrix by a 1D vector with incompatible inner dimensions is unclear, failing to specify which dimensions are mismatched. This ambiguity complicates debugging and fixing shape mismatch errors for developers.
• issues/110513

• Incorrect resource input classification causing performance degradation: A bug in auto_control_deps_utils.py causes unsorted _read_only_resource_inputs to be misclassified as read-write, leading to redundant control dependencies. This misclassification negatively impacts performance by introducing unnecessary overhead.
• issues/110516

• Lack of optimized kernel for depthwise convolution on Xtensa: The absence of an optimized depthwise convolution kernel for Xtensa with 16-bit activations and 8-bit weights forces fallback to a less efficient reference kernel. This results in suboptimal performance on supported hardware.
• issues/110532

• Heap corruption and crashes in XLA mode with zero-sized slice outputs: Using tf.slice to produce zero-sized outputs from dynamically-shaped inputs within jit_compile=True functions causes heap corruption and crashes in XLA mode, although eager mode executes successfully. This bug leads to instability and unexpected termination during compilation.
• issues/110789

• Improper error handling causing process aborts on invalid input specifications: Several operations, including tf.linalg.diag with invalid band specs on rank-1 tensors and tf.raw_ops.AvgPool3DGrad with non-5D gradients in MKL, cause C++ CHECK failures and process aborts instead of raising proper Python exceptions. This improper error handling results in abrupt termination rather than manageable error reporting.
• issues/110796, issues/110797

• XLA JIT compilation failures due to invalid dimension handling in StackV2 and tf.linalg.qr: Using StackV2 operations followed by tf.linalg.qr within XLA JIT compiled functions triggers C++ CHECK failures and process aborts caused by invalid dimension numbers during compilation. This issue disrupts the use of certain tensor operations in optimized compilation paths.
• issues/110798

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: 10

Key Open Pull Requests

1. [RFC] Improve symbol visibility for python libraries: This pull request aims to improve symbol visibility for Python libraries in TensorFlow by consolidating multiple shared libraries into a single monolithic .so file, enabling wrap_py_init by default to minimize exported symbols, localizing symbols in custom operation libraries, fixing missing dependencies, and significantly reducing global symbol exports to prevent conflicts with other libraries while also decreasing the package size.

• URL: pull/110502

• Associated Commits: d52d4, 4e8a0, d9b42, f9b94, 20e9f, fd80f, 5b310

2. Fix SparseConcat crash on integer overflow in shape computation: This pull request fixes a fatal crash in the SparseConcat operation caused by integer overflow during shape computation by adding explicit overflow validation, replacing unchecked dimension growth with safer methods, preventing fatal aborts, and including a regression test to cover large-shape overflow scenarios.

• URL: pull/110622

• Associated Commits: 44840, 83803, dcae9, b47ac

3. added same size validation for var and accum_update in ApplyAdadeltaOp. Fixes #104078: This pull request adds a validation check in the ApplyAdadeltaOp kernel to ensure that the input variables var and accum_update are of the same size, addressing a potential Null Pointer Dereference vulnerability on the GPU.

• URL: pull/110559

• Associated Commits: 99dfe

Other Open Pull Requests

• Resource input handling fixes: Multiple pull requests address logical flaws in the handling and categorization of _read_only_resource_inputs by replacing incorrect pointer-based scans with set-based lookups and ensuring the inputs are sorted before processing. These changes prevent misclassification of read-only resources as read-write and eliminate spurious control dependencies, improving consistency in resource input classification.
• pull/110665, pull/110822

• Kernel and operation validation improvements: Pull requests fix critical runtime errors by adding checks to CUDA kernels to prevent negative bin values and by adding validation guards in MatrixDiagOp to avoid CHECK failures when using band specifications with rank-1 diagonal inputs. These fixes prevent GPU out-of-bounds memory access and convert hard C++ aborts into clean Python errors, improving stability and error reporting.
• pull/110681, pull/110850

• Build and dependency updates: Some pull requests update build targets and dependencies, including adding the absl::log library to TensorFlow Lite for enhanced logging and updating the lit PyPI target reference to support bzlmod integration. These changes improve build system integration and logging capabilities.
• pull/110784, pull/110787

• Shape utility refactoring: One pull request refactors ShapeUtil.cc to improve memory safety and reliability by replacing manual array lookups with size checks, enhancing overflow detection, refining error handling to prevent crashes, and updating layout assignments for better compatibility with modern XLA standards. This refactor enhances robustness and maintainability of shape handling code.
• pull/110833

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. fix: tf.data debug mode breaks dataset.save() due to shard_func returning None: This pull request addresses a bug in TensorFlow's tf.data debug mode where calling dataset.save() fails due to the shard_func returning None, by modifying the dummy shard function to return a valid tensor value (initially np.int64(0), later changed to a plain int) to prevent a TypeError during tracing.

• URL: pull/110735

• Associated Commits: ebb55, 3aba0, bd19a

• Associated Commits: ebb55, 3aba0, bd19a

2. [mlir][tosa] Add README announcing migration: This pull request adds a README file to the MLIR TOSA directory that announces the migration and redirects users to the new location for the development of TensorFlow to TOSA and TensorFlow Lite to TOSA legalizations.

• URL: pull/110523

• Associated Commits: d69fa

• Associated Commits: d69fa

3. Remove TensorBoard as hard dependency from r2.21: This pull request removes TensorBoard as a mandatory install-time dependency from the TensorFlow r2.21 release, making it an optional package that users must explicitly install if needed, and updates the build and lockfiles accordingly to reflect this change.

• URL: pull/110709

• Associated Commits: 691d9

• Associated Commits: 691d9

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