Provide version bumps for 26.08

CMakeLists.txt

@@ -80,7 +80,7 @@ else()
   # - Prepare rapids-cmake -----------------------------------------------------
   set(rapids-cmake-version 26.08)
   file(DOWNLOAD
-    https://raw.githubusercontent.com/rapidsai/rapids-cmake/691ccac03770627311b271e9dc5820f6ea2b6346/RAPIDS.cmake
+    https://raw.githubusercontent.com/rapidsai/rapids-cmake/refs/tags/v26.08.00/RAPIDS.cmake
       ${CMAKE_BINARY_DIR}/RAPIDS.cmake)
   include(${CMAKE_BINARY_DIR}/RAPIDS.cmake)
   include(rapids-cmake)

cmake/thirdparty/get_cuml.cmake

@@ -55,6 +55,6 @@ endfunction()
 # CPM_raft_SOURCE=/path/to/local/raft
 find_and_configure_cuml(VERSION    26.08
                         FORK       rapidsai
-                        PINNED_TAG 91d972563e5f7994c3fa57d1cd7d5d6e4caefd4f
+                        PINNED_TAG 265b9da6a0e75dbef071a3168398b993a5ff6f0e
                         USE_TREELITE_STATIC ${TRITON_FIL_USE_TREELITE_STATIC}
                         )

cmake/thirdparty/get_nvforest.cmake

@@ -28,6 +28,7 @@ function(find_and_configure_nvforest)
             GIT_REPOSITORY https://github.com/${PKG_FORK}/nvforest.git
             GIT_TAG        ${PKG_PINNED_TAG}
             SOURCE_SUBDIR  cpp
+            PATCHES        "./patches/nvforest.patch"
             OPTIONS
               "BUILD_NVFOREST_TESTS OFF"
               "BUILD_SHARED_LIBS ON"
@@ -42,7 +43,7 @@ endfunction()
 
 find_and_configure_nvforest(VERSION    26.08
                             FORK       rapidsai
-                            PINNED_TAG dd49f1a906654f59f20023849b2f1cae820f601f
+                            PINNED_TAG cef3a50da0f74b0015876b9d6d424c86141898dc
                             USE_TREELITE_STATIC ${TRITON_FIL_USE_TREELITE_STATIC}
                             ENABLE_GPU ${TRITON_ENABLE_GPU}
                             )

cmake/thirdparty/patches/nvforest.patch

@@ -0,0 +1,40 @@
+From 60c8df5adef69291d4978b4631d58b2ab9ac6429 Mon Sep 17 00:00:00 2001
+From: Hyunsu Cho <phcho@nvidia.com>
+Date: Tue, 4 Aug 2026 03:06:43 -0700
+Subject: [PATCH] Check raft_handle_ for nullptr
+
+---
+ cpp/include/nvforest/handle.hpp | 12 +++++++++---
+ 1 file changed, 9 insertions(+), 3 deletions(-)
+
+diff --git a/cpp/include/nvforest/handle.hpp b/cpp/include/nvforest/handle.hpp
+index fd1f3bb..42066c8 100644
+--- a/cpp/include/nvforest/handle.hpp
++++ b/cpp/include/nvforest/handle.hpp
+@@ -18,14 +18,20 @@ struct handle_t {
+   handle_t(raft::handle_t const& raft_handle) : raft_handle_{&raft_handle} {}
+   auto get_next_usable_stream() const
+   {
++    if (!raft_handle_) { return cuda_stream{}; }
+     return cuda_stream{raft_handle_->get_next_usable_stream().value()};
+   }
+-  auto get_stream_pool_size() const { return raft_handle_->get_stream_pool_size(); }
++  auto get_stream_pool_size() const {
++    if (!raft_handle_) { return std::size_t{}; }
++    return raft_handle_->get_stream_pool_size();
++  }
+   auto get_usable_stream_count() const { return std::max(get_stream_pool_size(), std::size_t{1}); }
+   void synchronize() const
+   {
+-    raft_handle_->sync_stream_pool();
+-    raft_handle_->sync_stream();
++    if (raft_handle_) {
++      raft_handle_->sync_stream_pool();
++      raft_handle_->sync_stream();
++    }
+   }
+
+  private:
+--
+2.43.0
+

conda/environments/triton_benchmark.yml

@@ -4,8 +4,9 @@ channels:
   - nvidia
   - rapidsai
 dependencies:
-  - cuda-version=12.8
-  - cudf=26.06
+  - cuda-version=13.3
+  - cuda-python=13.3
+  - cudf=26.08
   - libcusolver
   - libcusparse
   - matplotlib
@@ -15,4 +16,3 @@ dependencies:
   - pip:
       - tritonclient[all]
       - protobuf
-      - git+https://github.com/rapidsai/rapids-triton.git@branch-25.08#subdirectory=python

conda/environments/triton_test.yml

@@ -6,9 +6,11 @@ channels:
 dependencies:
   - aws-sdk-cpp
   - clang-tools=19.1.7
-  - cuda-version=12.8
-  - cudf=26.06
-  - cuml=26.06
+  - cuda-version=13.3
+  - cuda-python=13.3
+  - cudf=26.08
+  - cuml=26.08
+  - nvforest=26.08
   - flake8
   - hypothesis
   - lightgbm
@@ -23,4 +25,3 @@ dependencies:
   - pip:
       - tritonclient[all]
       - protobuf
-      - git+https://github.com/rapidsai/rapids-triton.git@branch-25.08#subdirectory=python

conda/environments/triton_test_no_client.yml

@@ -6,9 +6,11 @@ channels:
 dependencies:
   - aws-sdk-cpp
   - clang-tools=19.1.7
-  - cuda-version=12.8
-  - cudf=26.06
-  - cuml=26.06
+  - cuda-version=13.3
+  - cuda-python=13.3
+  - cudf=26.08
+  - cuml=26.08
+  - nvforest=26.08
   - flake8
   - hypothesis
   - lightgbm

ops/Dockerfile

@@ -3,7 +3,7 @@
 # Arguments for controlling build details
 ###########################################################################################
 # Version of Triton to use
-ARG TRITON_VERSION=26.06
+ARG TRITON_VERSION=26.07
 # Base container image
 ARG BASE_IMAGE=nvcr.io/nvidia/tritonserver:${TRITON_VERSION}-py3
 # Whether or not to enable GPU build
@@ -64,8 +64,9 @@ RUN conda run --no-capture-output -n triton_test \
  && rm -r /sdk_install
 
 FROM wheel-install-${USE_CLIENT_WHEEL} as conda-test
+COPY ./rapids_triton_python /workspace/rapids_triton_python
 RUN conda run --no-capture-output -n triton_test \
-    pip install git+https://github.com/rapidsai/rapids-triton.git@v26.06.00#subdirectory=python
+    pip install /workspace/rapids_triton_python
 RUN conda-pack --ignore-missing-files -n triton_test -o /tmp/env.tar \
  && mkdir /conda/test/ \
  && cd /conda/test/ \
@@ -100,16 +101,16 @@ RUN apt-get update \
 # Stage immediately before building; useful for build iteration
 FROM base as build-prep
 
-RUN mkdir -p /rapids_triton/build /rapids_triton/install
+RUN mkdir -p /workspace/build /workspace/install
 
-COPY ./src /rapids_triton/src
-COPY ./CMakeLists.txt /rapids_triton
-COPY ./cmake /rapids_triton/cmake
+COPY ./src /workspace/src
+COPY ./CMakeLists.txt /workspace
+COPY ./cmake /workspace/cmake
 
 ARG BACKEND_NAME=fil
 ENV BACKEND_NAME=$BACKEND_NAME
 
-WORKDIR /rapids_triton/build
+WORKDIR /workspace/build
 
 # Remove potentially stale build artifacts
 RUN if [ -d /opt/tritonserver/backends/${BACKEND_NAME} ]; \
@@ -172,7 +173,7 @@ RUN source /conda/dev/bin/activate \
       -DTRITON_ENABLE_STATS="${TRITON_ENABLE_STATS}" \
       -DRAPIDS_DEPENDENCIES_VERSION="${RAPIDS_DEPENDENCIES_VERSION}" \
       -DTRITON_FIL_USE_TREELITE_STATIC="${TRITON_FIL_USE_TREELITE_STATIC}" \
-      -DCMAKE_INSTALL_PREFIX=/rapids_triton/install \
+      -DCMAKE_INSTALL_PREFIX=/workspace/install \
       ..;
 
 ENV CCACHE_DIR=/ccache

qa/L0_e2e/test_model.py

@@ -18,9 +18,17 @@
 from functools import lru_cache
 
 try:
-    import cuml
+    import nvforest
 except Exception:
-    cuml = None
+    nvforest = None
+try:
+    from cupy import asnumpy
+except Exception:
+
+    def asnumpy(x):
+        return x
+
+
 import numpy as np
 import pytest
 import treelite
@@ -199,20 +207,20 @@ def __init__(
                 with open(model_path, "rb") as pkl_file:
                     self._base_model = pickle.load(pkl_file)
             else:
-                self._base_model = cuml.ForestInference.load(
-                    model_path, is_classifier=is_classifier, model_type=model_format
+                self._base_model = nvforest.load_model(
+                    model_file=model_path, model_type=model_format
                 )
 
     def predict(self, inputs):
         if self.predict_proba:
-            result = self._base_model.predict_proba(inputs["input__0"])
+            result = asnumpy(self._base_model.predict_proba(inputs["input__0"]))
             if len(result.shape) == 1 or (
                 len(result.shape) == 2 and result.shape[1] == 1
             ):
                 result = result.reshape((-1, 1))
                 result = np.concatenate((1 - result, result), axis=1)
         else:
-            result = self._base_model.predict(inputs["input__0"])
+            result = asnumpy(self._base_model.predict(inputs["input__0"]))
         output = {"output__0": result.squeeze()}
         if self._run_treeshap:
             treeshap_result = self._xgb_model.predict(

rapids_triton_python/pyproject.toml

@@ -0,0 +1,3 @@
+[build-system]
+requires = ["setuptools", "wheel"]
+build-backend = "setuptools.build_meta"

rapids_triton_python/rapids_triton/__init__.py

@@ -0,0 +1,16 @@
+# Copyright (c) 2021, NVIDIA CORPORATION.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from rapids_triton.client import Client
+from rapids_triton.logging import logger