Bump tested versions to TensorFlow 2.21.0 / Keras 3.14.0 (#464)

* Bump tested versions to TensorFlow 2.21.0 / Keras 3.14.0

Update CI pin and the README's documented tested versions, and accept
the renamed Threshold layer config keys (threshold_value -> threshold,
value -> default_value) so models exported from Keras 3.14 load.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Bump TensorFlow to 2.21.0 in the perf-test Dockerfile

Also pin keras==3.14.0 explicitly so the perf-test image matches the
versions pinned in CI rather than relying on whatever Keras the
TensorFlow wheel happens to pull in.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Update perf-test Dockerfile to build TF 2.21 from source

TF 2.21's .bazelrc uses --enable_workspace, which requires Bazel 7+, so
bump the pinned bazel from 6.5.0 to 7.7.0 (matches TF 2.21's
.bazelversion). Also install xxd, which TF's XLA genrules now invoke
during the build.

Verified by running docker build of test/Dockerfile end-to-end: TF
wheel builds, save_application_examples.py converts all Keras
Application models, applications_performance loads and benchmarks them.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Bump CI runner image to ubuntu-24.04

keras==3.14.0 requires Python >=3.11; the ubuntu-22.04 runner ships
Python 3.10. Use ubuntu-24.04 (Python 3.12) so pip can resolve the
pinned version.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Add --break-system-packages to CI pip installs

Ubuntu 24.04 ships a PEP 668 externally-managed Python where system-wide
pip installs are blocked unless the flag is set, even from a sudo
shell. Required after the runner bump.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Drop redundant pip self-upgrade in CI

Ubuntu 24.04 ships pip 24.0, which already handles the TF 2.21 and
Keras 3.14 wheels. The self-upgrade step also runs into PEP 668's
RECORD-file check on apt-installed pip, so removing it both simplifies
setup and avoids that quirk.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Add --ignore-installed to CI pip install

Some apt-shipped Python packages on the ubuntu-24.04 runner (e.g.
python3-typing-extensions) lack RECORD files, so pip refuses to
uninstall them when a newer version is needed. Tell pip to overlay
the requested versions without uninstalling, mirroring how other CI
runners on PEP 668 systems handle this.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: Dobiasd

.github/workflows/ci.yml

@@ -4,16 +4,15 @@ on: [workflow_dispatch, push, pull_request]
 
 jobs:
   build_gcc:
-    runs-on: ubuntu-22.04
+    runs-on: ubuntu-24.04
     steps:
       - uses: actions/checkout@main
       - name: Setup
         run: |
           # system libs
           sudo apt-get install libblas-dev liblapack-dev libatlas-base-dev gfortran
           # python libs
-          sudo pip3 install --upgrade pip
-          sudo pip3 install numpy scipy h5py "tensorflow==2.18.0" "keras==3.8.0"
+          sudo pip3 install --break-system-packages --ignore-installed numpy scipy h5py "tensorflow==2.21.0" "keras==3.14.0"
           echo "Python version:"
           python3 --version
           echo "Version numbers of TensorFlow and Keras:"

README.md

@@ -137,8 +137,8 @@ Requirements and Installation
 
 - A **C++14**-compatible compiler: Compilers from these versions on are fine: GCC 4.9, Clang 3.7 (libc++ 3.7) and Visual C++ 2015
 - Python 3.9 or higher
-- TensorFlow 2.18.0
-- Keras 3.8.0
+- TensorFlow 2.21.0
+- Keras 3.14.0
 
 (These are the tested versions, but somewhat older ones might work too.)
 

include/fdeep/import_model.hpp

@@ -1037,8 +1037,14 @@ namespace internal {
         const get_param_f&, const nlohmann::json& data,
         const std::string& name)
     {
-        float_type threshold = data["config"]["threshold_value"];
-        float_type default_value = data["config"]["value"];
+        // Keras renamed these config keys: threshold_value -> threshold, value -> default_value.
+        const auto& config = data["config"];
+        float_type threshold = json_obj_has_member(config, "threshold")
+            ? static_cast<float_type>(config["threshold"])
+            : static_cast<float_type>(config["threshold_value"]);
+        float_type default_value = json_obj_has_member(config, "default_value")
+            ? static_cast<float_type>(config["default_value"])
+            : static_cast<float_type>(config["value"]);
         return std::make_shared<threshold_layer>(name, threshold, default_value);
     }
 

test/Dockerfile

@@ -6,19 +6,20 @@ FROM ubuntu:24.04
 RUN apt-get update -y --fix-missing
 ENV DEBIAN_FRONTEND=noninteractive
 RUN TZ=Etc/UTC apt-get -y install tzdata
-RUN apt-get install -y build-essential apt-utils cmake python3 python3-pip git llvm-17 clang-17 apt-transport-https curl gnupg patchelf
+RUN apt-get install -y build-essential apt-utils cmake python3 python3-pip git llvm-17 clang-17 apt-transport-https curl gnupg patchelf xxd
 
 RUN curl -fsSL https://bazel.build/bazel-release.pub.gpg | gpg --dearmor >bazel-archive-keyring.gpg
 RUN mv bazel-archive-keyring.gpg /usr/share/keyrings
 RUN echo "deb [arch=amd64 signed-by=/usr/share/keyrings/bazel-archive-keyring.gpg] https://storage.googleapis.com/bazel-apt stable jdk1.8" | tee /etc/apt/sources.list.d/bazel.list
-RUN apt-get update -y && apt-get install -y bazel-6.5.0
-RUN ln -s /usr/bin/bazel-6.5.0 /usr/bin/bazel
+RUN apt-get update -y && apt-get install -y bazel-7.7.0
+RUN ln -s /usr/bin/bazel-7.7.0 /usr/bin/bazel
 
-RUN git clone -b 'v2.18.0' --single-branch --depth 1 https://github.com/tensorflow/tensorflow.git
+RUN git clone -b 'v2.21.0' --single-branch --depth 1 https://github.com/tensorflow/tensorflow.git
 WORKDIR /tensorflow
 RUN ./configure
 RUN bazel build -c opt --copt=-march=native --jobs 16 --local_ram_resources=HOST_RAM*.3 -c opt //tensorflow/tools/pip_package:wheel --repo_env=WHEEL_NAME=tensorflow_cpu
-RUN pip install --break-system-packages bazel-bin/tensorflow/tools/pip_package/wheel_house/tensorflow_cpu-2.18.0-cp312-cp312-linux_x86_64.whl
+RUN pip install --break-system-packages bazel-bin/tensorflow/tools/pip_package/wheel_house/tensorflow_cpu-2.21.0-cp312-cp312-linux_x86_64.whl
+RUN pip install --break-system-packages "keras==3.14.0"
 WORKDIR /
 
 RUN git clone -b 'v0.2.24' --single-branch --depth 1 https://github.com/Dobiasd/FunctionalPlus && cd FunctionalPlus && mkdir -p build && cd build && cmake .. && make && make install