samples: refresh PyTorch, TensorFlow, and Pillow requirements

samples/cuda_core/blurImageUnifiedMemory/README.md

@@ -108,7 +108,7 @@ When returning a zero-copy view, the caller must close the buffers after use (e.
 - `cuda-python` package (13.0.0+)
 - `cuda-core` package (>=1.0.0)
 - `numpy` package (>=2.3.2)
-- `pillow` package (10.0.0+)
+- `pillow` package (13.0.0+)
 
 ### Platform Support:
 

samples/cuda_core/blurImageUnifiedMemory/blurImageUnifiedMemory.py

@@ -25,7 +25,7 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 # /// script
-# dependencies = ["cuda-python>=13.0.0", "cuda-core>=1.0.0", "numpy>=2.3.2", "pillow>=10.0.0"]
+# dependencies = ["cuda-python>=13.0.0", "cuda-core>=1.0.0", "numpy>=2.3.2", "pillow>=13.0.0"]
 # ///
 
 """

samples/cuda_core/blurImageUnifiedMemory/requirements.txt

@@ -3,4 +3,4 @@
 cuda-python>=13.0.0
 cuda-core>=1.0.0
 numpy>=2.3.2
-pillow>=10.0.0
+pillow>=13.0.0

samples/cuda_core/customPyTorchKernel/README.md

@@ -9,7 +9,7 @@ This sample demonstrates how to add a custom GPU operation to PyTorch using the
 - NVIDIA GPU with Compute Capability 7.0+
 - CUDA Toolkit 13.0+
 - Python 3.11+
-- PyTorch 2.0+
+- PyTorch 2.14.1+
 - cuda-python >= 13.0.0
 - cuda-core >=1.0.0
 
@@ -23,10 +23,10 @@ pip install -r requirements.txt
 **Windows users:** The default `torch` wheel on PyPI for Windows is CPU-only and will cause `torch.cuda.is_available()` to return `False`. Install a CUDA-enabled build from PyTorch's wheel index *before* (or after) the command above:
 
 ```bash
-pip install torch --index-url https://download.pytorch.org/whl/cu128
+pip install "torch>=2.14.1" --index-url https://download.pytorch.org/whl/cu130
 ```
 
-Replace `cu128` with the wheel suffix matching your installed CUDA driver (e.g. `cu121`, `cu124`, `cu126`, `cu128`). The driver's CUDA version must be >= the wheel's bundled runtime.
+Use the [PyTorch installation selector](https://pytorch.org/get-started/locally/) to choose a supported wheel for your platform. The driver's CUDA version must be >= the wheel's bundled runtime.
 
 ## How to Run
 

samples/cuda_core/customPyTorchKernel/customPyTorchKernel.py

@@ -25,7 +25,7 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 # /// script
-# dependencies = ["cuda-python>=13.0.0", "cuda-core>=1.0.0", "torch>=2.0.0"]
+# dependencies = ["cuda-python>=13.0.0", "cuda-core>=1.0.0", "torch>=2.14.1"]
 # ///
 
 """

samples/cuda_core/customPyTorchKernel/requirements.txt

@@ -3,8 +3,8 @@
 # NOTE: On Windows, the default `torch` wheel from PyPI is CPU-only and the
 # sample will fail with "Torch not compiled with CUDA enabled". Install a
 # CUDA-enabled torch from PyTorch's wheel index first (see README.md):
-#   pip install torch --index-url https://download.pytorch.org/whl/cu128
+#   pip install "torch>=2.14.1" --index-url https://download.pytorch.org/whl/cu130
 
 cuda-python>=13.0.0
 cuda-core>=1.0.0
-torch>=2.0.0
+torch>=2.14.1

samples/cuda_core/customTensorFlowKernel/README.md

@@ -20,7 +20,7 @@ Learn how to add a custom GPU operation to TensorFlow using `cuda.core` with `tf
 - NVIDIA GPU with Compute Capability 7.0+
 - CUDA Toolkit 13.0+
 - Python 3.11+
-- TensorFlow 2.10+
+- TensorFlow 2.21+
 - cuda-python >= 13.0.0
 - cuda-core >=1.0.0 (required for LEGACY_DEFAULT_STREAM)
 - numpy >= 2.3.2

samples/cuda_core/customTensorFlowKernel/customTensorFlowKernel.py

@@ -25,7 +25,7 @@
 # OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
 # /// script
-# dependencies = ["numpy>=2.3.2", "tensorflow>=2.10.0", "cupy-cuda13x>=14.0.0", "cuda-python>=13.0.0", "cuda-core>=1.0.0"]
+# dependencies = ["numpy>=2.3.2", "tensorflow>=2.21.0", "cupy-cuda13x>=14.0.0", "cuda-python>=13.0.0", "cuda-core>=1.0.0"]
 # ///
 
 """

samples/cuda_core/customTensorFlowKernel/requirements.txt

@@ -2,13 +2,13 @@
 #
 # This sample demonstrates a cuda.core-first approach:
 # - cuda-python and cuda-core: Primary GPU management (device, stream, kernel)
-# - TensorFlow 2.10+: Deep learning framework (tf.py_function, tf.custom_gradient)
+# - TensorFlow 2.21+: Deep learning framework (tf.py_function, tf.custom_gradient)
 # - CuPy: Internal helper for device pointer access only
 #
 # Note: cuda-core>=1.0.0 is required for LEGACY_DEFAULT_STREAM constant
 
 numpy>=2.3.2
-tensorflow>=2.10.0
+tensorflow>=2.21.0
 cupy-cuda13x>=14.0.0
 cuda-python>=13.0.0
 cuda-core>=1.0.0