Forward merge release/26.06 into main

.gitignore

@@ -36,11 +36,12 @@ _skbuild/
 junit-*.xml
 
 
-## files pickled in notebook when ran during python docstring generation
+## files generated by notebooks during docs builds
 docs/source/*.model
 docs/source/*.pkl
 docs/source/*.tl
 docs/source/*.joblib
+docs/source/**/*.onnx
 
 ## autosummary-generated API reference stubs
 docs/source/api/generated/

ci/build_wheel_cuml.sh

@@ -20,16 +20,17 @@ LIBCUML_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcuml_${RAPIDS_PY_CUDA_SUFFIX}" rap
 echo "libcuml-${RAPIDS_PY_CUDA_SUFFIX} @ file://$(echo "${LIBCUML_WHEELHOUSE}"/libcuml_*.whl)" >> "${PIP_CONSTRAINT}"
 
 EXCLUDE_ARGS=(
-  --exclude "libcuml.so"
-  --exclude "libcuvs.so"
-  --exclude "libraft.so"
   --exclude "libcublas.so.*"
   --exclude "libcublasLt.so.*"
   --exclude "libcufft.so.*"
+  --exclude "libcuml.so"
   --exclude "libcurand.so.*"
   --exclude "libcusolver.so.*"
   --exclude "libcusparse.so.*"
+  --exclude "libcuvs.so"
+  --exclude "libnvforest++.so"
   --exclude "libnvJitLink.so.*"
+  --exclude "libraft.so"
   --exclude "librapids_logger.so"
   --exclude "librmm.so"
 )

ci/build_wheel_libcuml.sh

@@ -29,17 +29,18 @@ rapids-pip-retry install \
 export PIP_NO_BUILD_ISOLATION=0
 
 EXCLUDE_ARGS=(
-  --exclude "libraft.so"
   --exclude "libcublas.so.*"
   --exclude "libcublasLt.so.*"
   --exclude "libcufft.so.*"
   --exclude "libcurand.so.*"
   --exclude "libcusolver.so.*"
   --exclude "libcusparse.so.*"
+  --exclude "libnccl.so.*"
+  --exclude "libnvforest++.so"
   --exclude "libnvJitLink.so.*"
+  --exclude "libraft.so"
   --exclude "librapids_logger.so"
   --exclude "librmm.so"
-  --exclude "libnccl.so.*"
 )
 
 export SKBUILD_CMAKE_ARGS="-DDISABLE_DEPRECATION_WARNINGS=ON;-DCUML_USE_CUVS_STATIC=ON"

ci/release/update-version.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2019-2025, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2019-2026, NVIDIA CORPORATION.
 # SPDX-License-Identifier: Apache-2.0
 ########################
 # cuML Version Updater #
@@ -106,9 +106,11 @@ DEPENDENCIES=(
   libcuml
   libcuml-tests
   libcuvs
+  libnvforest
   libraft-headers
   libraft
   librmm
+  nvforest
   pylibraft
   raft-dask
   rapids-dask-dependency

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

@@ -37,6 +37,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.8.*,>=0.0.0a0
+- libnvforest==26.8.*,>=0.0.0a0
 - libopenblas<=0.3.30
 - libraft==26.8.*,>=0.0.0a0
 - librmm==26.8.*,>=0.0.0a0
@@ -47,9 +48,10 @@ dependencies:
 - nltk
 - numba-cuda>=0.22.2,<0.29.0
 - numba>=0.60.0,<0.65.0
-- numpy>=1.23,<3.0
+- numpy>=1.26,<3.0
 - numpydoc
 - numpydoc<1.9
+- nvforest==26.8.*,>=0.0.0a0
 - nvidia-ml-py>=12
 - onnxruntime
 - packaging

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

@@ -37,6 +37,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.8.*,>=0.0.0a0
+- libnvforest==26.8.*,>=0.0.0a0
 - libraft==26.8.*,>=0.0.0a0
 - librmm==26.8.*,>=0.0.0a0
 - lightgbm
@@ -46,9 +47,10 @@ dependencies:
 - nltk
 - numba-cuda>=0.22.2,<0.29.0
 - numba>=0.60.0,<0.65.0
-- numpy>=1.23,<3.0
+- numpy>=1.26,<3.0
 - numpydoc
 - numpydoc<1.9
+- nvforest==26.8.*,>=0.0.0a0
 - nvidia-ml-py>=12
 - onnxruntime
 - packaging

conda/environments/all_cuda-132_arch-aarch64.yaml

@@ -37,6 +37,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.8.*,>=0.0.0a0
+- libnvforest==26.8.*,>=0.0.0a0
 - libopenblas<=0.3.30
 - libraft==26.8.*,>=0.0.0a0
 - librmm==26.8.*,>=0.0.0a0
@@ -47,9 +48,10 @@ dependencies:
 - nltk
 - numba-cuda>=0.22.2,<0.29.0
 - numba>=0.60.0,<0.65.0
-- numpy>=1.23,<3.0
+- numpy>=1.26,<3.0
 - numpydoc
 - numpydoc<1.9
+- nvforest==26.8.*,>=0.0.0a0
 - nvidia-ml-py>=12
 - onnxruntime
 - packaging

conda/environments/all_cuda-132_arch-x86_64.yaml

@@ -37,6 +37,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.8.*,>=0.0.0a0
+- libnvforest==26.8.*,>=0.0.0a0
 - libraft==26.8.*,>=0.0.0a0
 - librmm==26.8.*,>=0.0.0a0
 - lightgbm
@@ -46,9 +47,10 @@ dependencies:
 - nltk
 - numba-cuda>=0.22.2,<0.29.0
 - numba>=0.60.0,<0.65.0
-- numpy>=1.23,<3.0
+- numpy>=1.26,<3.0
 - numpydoc
 - numpydoc<1.9
+- nvforest==26.8.*,>=0.0.0a0
 - nvidia-ml-py>=12
 - onnxruntime
 - packaging

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

@@ -21,6 +21,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.8.*,>=0.0.0a0
+- libnvforest==26.8.*,>=0.0.0a0
 - libraft-headers==26.8.*,>=0.0.0a0
 - librmm==26.8.*,>=0.0.0a0
 - llvm-openmp==20.1.8

conda/environments/clang_tidy_cuda-132_arch-x86_64.yaml

@@ -21,6 +21,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.8.*,>=0.0.0a0
+- libnvforest==26.8.*,>=0.0.0a0
 - libraft-headers==26.8.*,>=0.0.0a0
 - librmm==26.8.*,>=0.0.0a0
 - llvm-openmp==20.1.8

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

@@ -19,6 +19,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.8.*,>=0.0.0a0
+- libnvforest==26.8.*,>=0.0.0a0
 - libraft-headers==26.8.*,>=0.0.0a0
 - librmm==26.8.*,>=0.0.0a0
 - ninja

conda/environments/cpp_all_cuda-132_arch-x86_64.yaml

@@ -19,6 +19,7 @@ dependencies:
 - libcusolver-dev
 - libcusparse-dev
 - libcuvs==26.8.*,>=0.0.0a0
+- libnvforest==26.8.*,>=0.0.0a0
 - libraft-headers==26.8.*,>=0.0.0a0
 - librmm==26.8.*,>=0.0.0a0
 - ninja

conda/recipes/cuml/recipe.yaml

@@ -80,6 +80,7 @@ requirements:
     - cudf =${{ minor_version }}
     - cython >=3.2.2
     - libcuml =${{ version }}
+    - nvforest =${{ minor_version }}
     - pip
     - pylibraft =${{ minor_version }}
     - python =${{ py_abi_min }}
@@ -100,7 +101,8 @@ requirements:
     - libcuml =${{ version }}
     - numba >=0.60.0,<0.65.0
     - numba-cuda >=0.22.2,<0.29.0
-    - numpy >=1.23,<3.0
+    - numpy >=1.26,<3.0
+    - nvforest =${{ minor_version }}
     - scikit-learn >=1.4
     - scipy >=1.14.0
     - packaging

conda/recipes/libcuml/recipe.yaml

@@ -106,6 +106,7 @@ outputs:
         - cuda-version =${{ cuda_version }}
         - rapids-logger =0.2
         - libcuvs =${{ minor_version }}
+        - libnvforest =${{ minor_version }}
         - librmm =${{ minor_version }}
         - treelite ${{ treelite_version }}
         - cuda-cudart-dev
@@ -122,6 +123,7 @@ outputs:
         - libcusolver
         - libcusparse
         - libcuvs =${{ minor_version }}
+        - libnvforest =${{ minor_version }}
         - librmm =${{ minor_version }}
         - rapids-logger =0.2
         - treelite ${{ treelite_version }}

cpp/CMakeLists.txt

@@ -70,8 +70,12 @@ option(CUML_COMPILE_DYNAMIC_ONLY "Only build the shared library and skip the sta
 # dependent library.
 option(CUML_EXCLUDE_RAFT_FROM_ALL "Exclude RAFT targets from cuML's 'all' target" OFF)
 option(CUML_EXCLUDE_TREELITE_FROM_ALL "Exclude Treelite targets from cuML's 'all' target" OFF)
+option(CUML_EXCLUDE_NVFOREST_FROM_ALL "Exclude nvForest targets from cuML's 'all' target" OFF)
 option(CUML_RAFT_CLONE_ON_PIN "Explicitly clone RAFT branch when pinned to non-feature branch" ON)
 option(CUML_CUVS_CLONE_ON_PIN "Explicitly clone CUVS branch when pinned to non-feature branch" ON)
+option(CUML_NVFOREST_CLONE_ON_PIN
+       "Explicitly clone nvForest branch when pinned to non-feature branch" ON
+)
 
 message(VERBOSE "CUML_CPP: Building libcuml shared library: ${BUILD_CUML_CPP_LIBRARY}")
 message(VERBOSE "CUML_CPP: Building cuML algorithm tests: ${BUILD_CUML_TESTS}")
@@ -270,6 +274,11 @@ if(LINK_TREELITE)
   include(cmake/thirdparty/get_treelite.cmake)
 endif()
 
+if(LINK_NVFOREST)
+  include(cmake/thirdparty/get_nvforest.cmake)
+  set(CUML_NVFOREST_TARGET nvforest::nvforest++)
+endif()
+
 if(all_algo OR treeshap_algo)
   include(cmake/thirdparty/get_gputreeshap.cmake)
 endif()
@@ -367,42 +376,6 @@ if(BUILD_CUML_CPP_LIBRARY)
     target_sources(cuml_objs PRIVATE src/explainer/tree_shap.cu)
   endif()
 
-  # FIL components
-  if(all_algo OR fil_algo)
-    if(CUML_ENABLE_GPU)
-      target_sources(
-        cuml_objs
-        PRIVATE src/fil/infer0.cu
-                src/fil/infer1.cu
-                src/fil/infer2.cu
-                src/fil/infer3.cu
-                src/fil/infer4.cu
-                src/fil/infer5.cu
-                src/fil/infer6.cu
-                src/fil/infer7.cu
-                src/fil/infer8.cu
-                src/fil/infer9.cu
-                src/fil/infer10.cu
-                src/fil/infer11.cu
-      )
-    endif()
-    target_sources(
-      cuml_objs
-      PRIVATE src/fil/infer0.cpp
-              src/fil/infer1.cpp
-              src/fil/infer2.cpp
-              src/fil/infer3.cpp
-              src/fil/infer4.cpp
-              src/fil/infer5.cpp
-              src/fil/infer6.cpp
-              src/fil/infer7.cpp
-              src/fil/infer8.cpp
-              src/fil/infer9.cpp
-              src/fil/infer10.cpp
-              src/fil/infer11.cpp
-    )
-  endif()
-
   # todo: organize linear models better
   if(all_algo
      OR linearregression_algo
@@ -596,6 +569,10 @@ if(BUILD_CUML_CPP_LIBRARY)
     copy_interface_excludes(INCLUDED_TARGET treelite::treelite_static TARGET cuml_objs)
   endif()
 
+  if(LINK_NVFOREST)
+    list(APPEND _cuml_cpp_public_libs ${CUML_NVFOREST_TARGET})
+  endif()
+
   # These are always private:
   list(
     APPEND

cpp/bench/CMakeLists.txt

@@ -1,6 +1,6 @@
 # =============================================================================
 # cmake-format: off
-# SPDX-FileCopyrightText: Copyright (c) 2019-2025, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2019-2026, NVIDIA CORPORATION.
 # SPDX-License-Identifier: Apache-2.0
 # cmake-format: on
 # =============================================================================
@@ -24,7 +24,6 @@ if(BUILD_CUML_BENCH)
     sg/svc.cu
     sg/svr.cu
     sg/umap.cu
-    sg/fil.cu
   )
   if(CUML_ENABLE_GPU)
     target_compile_definitions(${CUML_CPP_BENCH_TARGET} PUBLIC CUML_ENABLE_GPU)