use 'nvidia-ml-py' instead of 'pynvml', declare 'numba-cuda' dependency pins

conda/environments/all_cuda-129_arch-aarch64.yaml

@@ -12,10 +12,10 @@ dependencies:
 - cuda-cudart-dev
 - cuda-nvcc
 - cuda-profiler-api
-- cuda-python>=12.6.2,<13.0a0
+- cuda-python>=12.9.2,<13.0a0
 - cuda-version=12.9
 - cudf==25.10.*,>=0.0.0a0
-- cupy>=12.0.0
+- cupy>=13.6.0
 - cuvs==25.10.*,>=0.0.0a0
 - cxx-compiler
 - cython>=3.0.0
@@ -43,16 +43,17 @@ dependencies:
 - nbsphinx
 - ninja
 - nltk
-- numba>=0.59.1,<0.62.0a0
+- numba-cuda>=0.19.1,<0.20.0a0
+- numba>=0.60.0,<0.62.0a0
 - numpy>=1.23,<3.0a0
 - numpydoc
 - numpydoc<1.9
+- nvidia-ml-py>=12.560.30
 - packaging
 - pre-commit
 - pydata-sphinx-theme!=0.14.2
 - pylibraft==25.10.*,>=0.0.0a0
 - pynndescent
-- pynvml>=12.0.0,<13.0.0a0
 - pytest
 - pytest-benchmark
 - pytest-cases

conda/environments/all_cuda-129_arch-x86_64.yaml

@@ -12,10 +12,10 @@ dependencies:
 - cuda-cudart-dev
 - cuda-nvcc
 - cuda-profiler-api
-- cuda-python>=12.6.2,<13.0a0
+- cuda-python>=12.9.2,<13.0a0
 - cuda-version=12.9
 - cudf==25.10.*,>=0.0.0a0
-- cupy>=12.0.0
+- cupy>=13.6.0
 - cuvs==25.10.*,>=0.0.0a0
 - cxx-compiler
 - cython>=3.0.0
@@ -43,16 +43,17 @@ dependencies:
 - nbsphinx
 - ninja
 - nltk
-- numba>=0.59.1,<0.62.0a0
+- numba-cuda>=0.19.1,<0.20.0a0
+- numba>=0.60.0,<0.62.0a0
 - numpy>=1.23,<3.0a0
 - numpydoc
 - numpydoc<1.9
+- nvidia-ml-py>=12.560.30
 - packaging
 - pre-commit
 - pydata-sphinx-theme!=0.14.2
 - pylibraft==25.10.*,>=0.0.0a0
 - pynndescent
-- pynvml>=12.0.0,<13.0.0a0
 - pytest
 - pytest-benchmark
 - pytest-cases

conda/recipes/cuml/recipe.yaml

@@ -70,15 +70,18 @@ requirements:
     - scikit-build-core >=0.10.0
     - treelite ${{ treelite_version }}
     - cuda-cudart-dev
-    - cuda-python >=12.6.2,<13.0a0
+    - if: cuda_major == "12"
+      then: cuda-python >=12.9.2,<13.0a0
   run:
     - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="x") }}
     - cudf =${{ minor_version }}
-    - cupy >=12.0.0
+    - cupy >=13.6.0
     - dask-cudf =${{ minor_version }}
     - joblib >=0.11
     - libcuml =${{ version }}
     - libcumlprims =${{ minor_version }}
+    - numba >=0.60.0,<0.62.0a0
+    - numba-cuda >=0.19.1,<0.20.0a0
     - numpy >=1.23,<3.0a0
     - scikit-learn >=1.4
     - packaging
@@ -90,7 +93,8 @@ requirements:
     - rich
     - treelite ${{ treelite_version }}
     - cuda-cudart
-    - cuda-python >=12.6.2,<13.0a0
+    - if: cuda_major == "12"
+      then: cuda-python >=12.9.2,<13.0a0
   ignore_run_exports:
     by_name:
       - cuda-cudart

dependencies.yaml

@@ -13,11 +13,13 @@ files:
       - depends_on_cuda_python
       - depends_on_cupy
       - depends_on_cuvs
+      - depends_on_dask_cuda
       - depends_on_dask_cudf
       - depends_on_libcumlprims
       - depends_on_libcuvs
       - depends_on_libraft
       - depends_on_librmm
+      - depends_on_numba_cuda
       - depends_on_pylibraft
       - depends_on_raft_dask
       - depends_on_rapids_logger
@@ -41,11 +43,13 @@ files:
       - depends_on_cuda_python
       - depends_on_cupy
       - depends_on_cuvs
+      - depends_on_dask_cuda
       - depends_on_dask_cudf
       - depends_on_libcumlprims
       - depends_on_libcuvs
       - depends_on_libraft
       - depends_on_librmm
+      - depends_on_numba_cuda
       - depends_on_pylibraft
       - depends_on_raft_dask
       - depends_on_rapids_logger
@@ -119,7 +123,9 @@ files:
       - depends_on_cuml
       - depends_on_cupy
       - depends_on_cuvs
+      - depends_on_dask_cuda
       - depends_on_dask_cudf
+      - depends_on_numba_cuda
       - depends_on_pylibraft
       - depends_on_raft_dask
       - depends_on_rmm
@@ -161,8 +167,10 @@ files:
       - depends_on_cudf
       - depends_on_cupy
       - depends_on_cuvs
+      - depends_on_dask_cuda
       - depends_on_dask_cudf
       - depends_on_libcuml
+      - depends_on_numba_cuda
       - depends_on_pylibraft
       - depends_on_raft_dask
       - depends_on_rmm
@@ -289,9 +297,8 @@ dependencies:
     common:
       - output_types: [conda, requirements, pyproject]
         packages:
-          - dask-cuda==25.10.*,>=0.0.0a0
           - joblib>=0.11
-          - numba>=0.59.1,<0.62.0a0
+          - numba>=0.60.0,<0.62.0a0
           - &numpy numpy>=1.23,<3.0a0
           - scipy>=1.8.0
           - packaging
@@ -449,7 +456,9 @@ dependencies:
           - nltk
           # upstream sklearn docstring tests require numpydoc<1.9
           - numpydoc<1.9
-          - pynvml>=12.0.0,<13.0.0a0
+          # 'nvidia-ml-py' provides the 'pynvml' module, since v12.560.30
+          # ref: https://github.com/conda-forge/nvidia-ml-py-feedstock/pull/24
+          - nvidia-ml-py>=12.560.30
           - pyyaml
           - pytest
           - pytest-benchmark
@@ -495,7 +504,7 @@ dependencies:
           - matrix:
               cuda: "12.*"
             packages:
-              - cuda-python>=12.6.2,<13.0a0
+              - cuda-python>=12.9.2,<13.0a0
           - matrix:
             packages:
               - cuda-python
@@ -529,13 +538,16 @@ dependencies:
     common:
       - output_types: conda
         packages:
-          - cupy>=12.0.0
+          - cupy>=13.6.0
+    # NOTE: This is intentionally not broken into groups by a 'cuda_suffixed' selector like
+    #       other packages with -cu{nn}x suffixes in this file.
+    #       All RAPIDS wheel builds (including in devcontainers) expect cupy to be suffixed.
     specific:
       - output_types: [requirements, pyproject]
         matrices:
-          - matrix: # All CUDA 12 versions
+          - matrix:
             packages:
-              - cupy-cuda12x>=12.0.0
+              - cupy-cuda12x>=13.6.0
   depends_on_cuvs:
     common:
       - output_types: conda
@@ -557,6 +569,27 @@ dependencies:
           - matrix:
             packages:
               - *cuvs_unsuffixed
+  depends_on_dask_cuda:
+    common:
+      - output_types: conda
+        packages:
+          - &dask_cuda_unsuffixed dask-cuda==25.10.*,>=0.0.0a0
+      - output_types: requirements
+        packages:
+          # pip recognizes the index as a global option for the requirements.txt file
+          - --extra-index-url=https://pypi.nvidia.com
+          - --extra-index-url=https://pypi.anaconda.org/rapidsai-wheels-nightly/simple
+    specific:
+      - output_types: [requirements, pyproject]
+        matrices:
+          - matrix:
+              cuda: "12.*"
+              cuda_suffixed: "true"
+            packages:
+              - dask-cuda[cu12]==25.10.*,>=0.0.0a0
+          - matrix:
+            packages:
+              - *dask_cuda_unsuffixed
   depends_on_dask_cudf:
     common:
       - output_types: conda
@@ -664,6 +697,21 @@ dependencies:
             packages:
               - librmm-cu12==25.10.*,>=0.0.0a0
           - {matrix: null, packages: [*librmm_unsuffixed]}
+  depends_on_numba_cuda:
+    common:
+      - output_types: [conda]
+        packages:
+        - numba-cuda>=0.19.1,<0.20.0a0
+    specific:
+      - output_types: [requirements, pyproject]
+        matrices:
+          - matrix:
+              cuda: "12.*"
+            packages:
+              - numba-cuda[cu12]>=0.19.1,<0.20.0a0
+          - matrix:
+            packages:
+              - numba-cuda>=0.19.1,<0.20.0a0
   depends_on_pylibraft:
     common:
       - output_types: conda

docs/source/supported_versions.rst

@@ -14,7 +14,7 @@ The following dependencies are required for the cuML library:
 * **scikit-learn**: >=1.4
 * **scipy**: >=1.8.0
 * **numba**: >=0.59.1,<0.62.0a0
-* **cupy**: cupy-cuda12x>=12.0.0
+* **cupy**: cupy-cuda12x>=13.6.0 (CUDA 12)
 * **treelite**: ==4.4.1
 
 

python/cuml/README.md

@@ -55,13 +55,13 @@ cuML's convenience [development yaml files](https://github.com/rapidsai/cuml/tre
 To build cuML's Python package, the following dependencies are required:
 
 - cudatoolkit version corresponding to system CUDA toolkit
-- cython >= 0.29, < 0.30
+- cython >=3.0.0
 - numpy
-- cmake >= 3.14
+ cmake >=3.30.4
 - cudf version matching the cuML version
 - libcuml version matching the cuML version
 - libcuml={{ version }}
-- cupy>=7.8.0,<12.0.0a0
+- cupy >=13.6.0
 - joblib >=0.11
 
 Packages required for multigpu algorithms*:

python/cuml/pyproject.toml

@@ -93,13 +93,14 @@ requires-python = ">=3.10"
 dependencies = [
     "cuda-python",
     "cudf==25.10.*,>=0.0.0a0",
-    "cupy-cuda12x>=12.0.0",
+    "cupy-cuda12x>=13.6.0",
     "cuvs==25.10.*,>=0.0.0a0",
     "dask-cuda==25.10.*,>=0.0.0a0",
     "dask-cudf==25.10.*,>=0.0.0a0",
     "joblib>=0.11",
     "libcuml==25.10.*,>=0.0.0a0",
-    "numba>=0.59.1,<0.62.0a0",
+    "numba-cuda>=0.19.1,<0.20.0a0",
+    "numba>=0.60.0,<0.62.0a0",
     "numpy>=1.23,<3.0a0",
     "nvidia-cublas",
     "nvidia-cufft",
@@ -133,8 +134,8 @@ test = [
     "hypothesis>=6.0,<7",
     "nltk",
     "numpydoc<1.9",
+    "nvidia-ml-py>=12.560.30",
     "pynndescent",
-    "pynvml>=12.0.0,<13.0.0a0",
     "pytest",
     "pytest-benchmark",
     "pytest-cases",