dependabot[bot] opened a new pull request, #39687: URL: https://github.com/apache/beam/pull/39687
Bumps [keras](https://github.com/keras-team/keras) from 3.12.3 to 3.15.0. <details> <summary>Release notes</summary> <p><em>Sourced from <a href="https://github.com/keras-team/keras/releases";>keras's releases</a>.</em></p> <blockquote> <h2>v3.15.0</h2> <h2>Highlights</h2> <ul> <li><strong>Keras-to-Torch Export</strong>: New <code>export_torch</code> enables exporting Keras models to native PyTorch <code>nn.Module</code> format, along with LiteRT (TFLite) export support for the PyTorch backend.</li> <li><strong>Sliding Window Attention</strong>: Added <code>sliding_window</code> parameter to <code>MultiHeadAttention</code> and <code>GroupedQueryAttention</code> for efficient long-context attention.</li> <li><strong>Flash / Fused SDPA</strong>: Causal-only MHA/GQA now automatically dispatches to Flash Attention (cuDNN SDPA), and the manual attention path correctly applies causal masking.</li> <li><strong>Multi-Optimizer Training</strong>: New <code>MultiOptimizer</code> supports assigning different optimizers to sub-networks.</li> <li><strong>New Math Operations</strong>: Added <code>unique</code>, <code>pinv</code>, <code>matrix_rank</code>, <code>fabs</code>, <code>fmax</code>, <code>fmin</code>, <code>erfc</code>, <code>dsplit</code>, <code>percentile</code>, <code>nanpercentile</code>, <code>sobel_edges</code>, and <code>ssim</code> (structural similarity) to <code>keras.ops</code>.</li> <li><strong>Security Hardening</strong>: Comprehensive hardening of model reloading against HDF5 exploits, tar/zip traversal attacks, insecure deserialization.</li> </ul> <hr /> <h2>New Features and Operations</h2> <h3>Multi-Backend Operations</h3> <ul> <li><strong>New NumPy Operations</strong>: Added <code>unique</code>, <code>fabs</code>, <code>fmax</code>, <code>fmin</code>, <code>dsplit</code>, <code>erfc</code>, <code>percentile</code>, <code>nanpercentile</code> in <code>keras.ops.numpy</code>.</li> <li><strong>New Linear Algebra Operations</strong>: Added <code>pinv</code> (pseudo-inverse) and <code>matrix_rank</code> in <code>keras.ops.linalg</code>.</li> <li><strong>New Image Operations</strong>: Added <code>sobel_edges</code> for edge detection and <code>ssim</code> (structural similarity) in <code>keras.ops.image</code>.</li> <li><strong>Negative Axes in Transpose</strong>: <code>keras.ops.transpose</code> now supports negative axis values.</li> </ul> <h3>Layers and Attention</h3> <ul> <li><strong>Sliding Window Attention</strong>: <code>MultiHeadAttention</code> and <code>GroupedQueryAttention</code> layers support the <code>sliding_window</code> parameter for efficient long-sequence processing.</li> <li><strong>Flash Attention Engagement</strong>: Causal-only attention in MHA/GQA now uses Flash SDPA for significant speedups.</li> <li><strong>Fused Bidirectional LSTM/GRU</strong>: JAX backend now fuses Bidirectional LSTM into a single cuDNN call; fused bidirectional GRU added for Torch backend.</li> <li><strong>CTC Beam Search Decoder</strong>: Added CTC beam search decoding for the Torch backend.</li> </ul> <h3>Training and Optimizers</h3> <ul> <li><strong>MultiOptimizer</strong>: Supports training sub-networks with different optimizers.</li> <li><strong>SKLearn Classifier</strong>: Added <code>predict_proba</code> method to <code>SKLearnClassifier</code>.</li> </ul> <hr /> <h2>Export and Deployment</h2> <ul> <li><strong>Keras-to-Torch Export</strong>: Export Keras models to native PyTorch <code>nn.Module</code> via <code>model.export(..., format="torch")</code>.</li> <li><strong>LiteRT (TFLite) Export for PyTorch</strong>: Added LiteRT export support for models using the PyTorch backend.</li> <li><strong>LiteRT Compatibility Fix</strong>: Fixed LiteRT export for Keras 3 + TF 2.20 + Python 3.13.</li> <li><strong>ONNX Export</strong>: Support for dict/list inputs in Torch ONNX export; documented static input signature requirement for LiteRT PyTorch export.</li> </ul> <hr /> <h2>Distribution and Parallelism</h2> <ul> <li><strong>ModelParallel Improvements</strong>: Defined contiguous replica-group data shard ID convention; added distribution information (<code>num_processes</code>, <code>num_model_replicas</code>, <code>data_shard_id</code>).</li> <li><strong>Initializer Distribution Layout</strong>: Initializers can now handle the distribution layout directly with JAX.</li> <li><strong>TF Dataset Distribution</strong>: Refactored TF dataset distribution with centralized sharding routing; fixed data distribution for model training in JAX.</li> </ul> <hr /> <h2>OpenVINO Backend Support</h2> <!-- raw HTML omitted --> </blockquote> <p>... (truncated)</p> </details> <details> <summary>Commits</summary> <ul> <li><a href="https://github.com/keras-team/keras/commit/9d1bbf9aae730708cc6e0d7104860c1605c0e31b";><code>9d1bbf9</code></a> Add erfinv tests and TensorFlow CPU fallback for half precision (<a href="https://redirect.github.com/keras-team/keras/issues/23086";>#23086</a>)</li> <li><a href="https://github.com/keras-team/keras/commit/b34af30ad9898b4aadeccde6a7d0daa1702a5a59";><code>b34af30</code></a> Fix pad_sequences string dtype check (np.bytes_ instead of duplicate np.str_)...</li> <li><a href="https://github.com/keras-team/keras/commit/f512e932b4f93849dd99304dee7ddfea93a502cc";><code>f512e93</code></a> Fix wrong parameter names in ops.select and ops.argpartition docstrings (<a href="https://redirect.github.com/keras-team/keras/issues/23091";>#23091</a>)</li> <li><a href="https://github.com/keras-team/keras/commit/915ec7e3df4367053de6fed084337663ff8e92b8";><code>915ec7e</code></a> Support native Grouped-Query Attention (GQA) Key/Value head broadcast. (<a href="https://redirect.github.com/keras-team/keras/issues/23081";>#23081</a>)</li> <li><a href="https://github.com/keras-team/keras/commit/49d668dd973399b3379f9a509f9435decbc018e4";><code>49d668d</code></a> Add JAX multi-process distribution tests (<a href="https://redirect.github.com/keras-team/keras/issues/23105";>#23105</a>)</li> <li><a href="https://github.com/keras-team/keras/commit/b49fde7f913c71d6e5dbd2a796ef45e1499b49eb";><code>b49fde7</code></a> Refactor rematerialization logic into Operation and support more modes (<a href="https://redirect.github.com/keras-team/keras/issues/23107";>#23107</a>)</li> <li><a href="https://github.com/keras-team/keras/commit/e0bfa2dcc300ee3d1c81c8e885cbe32e20a2928d";><code>e0bfa2d</code></a> Add to the list of APIs that should not be part of a reloaded model. (<a href="https://redirect.github.com/keras-team/keras/issues/23115";>#23115</a>)</li> <li><a href="https://github.com/keras-team/keras/commit/69d7fd6bb63a29bb2b355807e30eaafd3cf9e782";><code>69d7fd6</code></a> Suppress spurious 'Skipping nested container' warning on freshly-saved files ...</li> <li><a href="https://github.com/keras-team/keras/commit/048d14274e24d6a08e344fc086f6a8585f445379";><code>048d142</code></a> Delete temp files after loading model from remote path. (<a href="https://redirect.github.com/keras-team/keras/issues/23113";>#23113</a>)</li> <li><a href="https://github.com/keras-team/keras/commit/46f5eac7f4ad16ecc3e3c29047a77befe778ce7c";><code>46f5eac</code></a> Bump GitHub Actions versions to the latest (<a href="https://redirect.github.com/keras-team/keras/issues/23104";>#23104</a>)</li> <li>Additional commits viewable in <a href="https://github.com/keras-team/keras/compare/v3.12.3...v3.15.0";>compare view</a></li> </ul> </details> <br /> [](https://docs.github.com/en/github/managing-security-vulnerabilities/about-dependabot-security-updates#about-compatibility-scores) Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting `@dependabot rebase`. [//]: # (dependabot-automerge-start) [//]: # (dependabot-automerge-end) --- <details> <summary>Dependabot commands and options</summary> <br /> You can trigger Dependabot actions by commenting on this PR: - `@dependabot rebase` will rebase this PR - `@dependabot recreate` will recreate this PR, overwriting any edits that have been made to it - `@dependabot show <dependency name> ignore conditions` will show all of the ignore conditions of the specified dependency - `@dependabot ignore this major version` will close this PR and stop Dependabot creating any more for this major version (unless you reopen the PR or upgrade to it yourself) - `@dependabot ignore this minor version` will close this PR and stop Dependabot creating any more for this minor version (unless you reopen the PR or upgrade to it yourself) - `@dependabot ignore this dependency` will close this PR and stop Dependabot creating any more for this dependency (unless you reopen the PR or upgrade to it yourself) You can disable automated security fix PRs for this repo from the [Security Alerts page](https://github.com/apache/beam/network/alerts). </details> -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
