build wheels with CUDA 13.0.x, test wheels against mix of CTK versions, drop CUDA math libraries dependencies

[jameslamb]

Contributes to rapidsai/build-planning#257

• builds CUDA 13 wheels with the 13.0 CTK

Contributes to rapidsai/build-planning#256

• updates wheel tests to cover a range of CTK versions (we previously, accidentally, were only testing the latest 12.x and 13.x)

Drops unnecessary CTK dependencies

• drops dependencies on cuBLAS, cuFFT, cuRAND, cuSOLVER, and cuSPARSE
  • nvforest doesn't depend directly on any of these, I suspect they were all just copied over from cuML
• removes related code and configuration

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[jameslamb]

/ok to test

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 68b7043b-a86d-4284-86ab-05baa7f1b989

📥 Commits

Reviewing files that changed from the base of the PR and between 705406e and a70f15f.

📒 Files selected for processing (19)

• .devcontainer/cuda12.9-pip/devcontainer.json
• .devcontainer/cuda13.1-pip/devcontainer.json
• .github/workflows/build.yaml
• .github/workflows/pr.yaml
• ci/test_wheel.sh
• conda/environments/all_cuda-129_arch-aarch64.yaml
• conda/environments/all_cuda-129_arch-x86_64.yaml
• conda/environments/all_cuda-131_arch-aarch64.yaml
• conda/environments/all_cuda-131_arch-x86_64.yaml
• conda/environments/clang_tidy_cuda-129_arch-x86_64.yaml
• conda/environments/clang_tidy_cuda-131_arch-x86_64.yaml
• conda/environments/cpp_all_cuda-129_arch-x86_64.yaml
• conda/environments/cpp_all_cuda-131_arch-x86_64.yaml
• conda/recipes/libnvforest/recipe.yaml
• cpp/CMakeLists.txt
• dependencies.yaml
• python/libnvforest/CMakeLists.txt
• python/libnvforest/pyproject.toml
• python/nvforest/pyproject.toml

💤 Files with no reviewable changes (12)

• conda/environments/cpp_all_cuda-129_arch-x86_64.yaml
• conda/environments/all_cuda-131_arch-x86_64.yaml
• python/libnvforest/CMakeLists.txt
• conda/environments/all_cuda-129_arch-x86_64.yaml
• python/nvforest/pyproject.toml
• conda/environments/clang_tidy_cuda-131_arch-x86_64.yaml
• conda/environments/all_cuda-131_arch-aarch64.yaml
• conda/recipes/libnvforest/recipe.yaml
• python/libnvforest/pyproject.toml
• conda/environments/all_cuda-129_arch-aarch64.yaml
• conda/environments/clang_tidy_cuda-129_arch-x86_64.yaml
• conda/environments/cpp_all_cuda-131_arch-x86_64.yaml

---

📝 Walkthrough

Summary by CodeRabbit

• Build & Infrastructure

  • Updated build workflows and development container configurations for improved efficiency
  • Streamlined constraint handling in test workflows using environment variables

• Dependencies

  • Removed optional CUDA library development packages from build and runtime environments
  • Simplified CUDA dependency management across multiple environment configurations
  • Reduced CUDA toolkit version constraints in Python packages for greater compatibility

Walkthrough

This PR disables CUDA library components (cuBLAS, cuSOLVER, cuRAND, cuSPARSE) in devcontainers, removes CUDA library dev packages from Conda environment files, updates GitHub workflow references, refactors CUDA version management in dependencies, simplifies CMake CUDA handling, and removes explicit CUDA toolkit version constraints from Python projects.

Changes

Cohort / File(s)	Summary
DevContainer CUDA Configuration .devcontainer/cuda12.9-pip/devcontainer.json, .devcontainer/cuda13.1-pip/devcontainer.json	Disabled CUDA feature installation flags for cuBLAS, cuSOLVER, cuRAND, and cuSPARSE by changing from true to false.
GitHub Workflows .github/workflows/build.yaml, .github/workflows/pr.yaml	Updated reusable workflow references from release/26.04 to build-wheels-old-ctk for wheel-build jobs.
Test Infrastructure ci/test_wheel.sh	Replaced hard-coded constraint file references with environment-provided PIP_CONSTRAINT variable and added --prefer-binary flag to installation steps.
Conda Environment Definitions conda/environments/all_cuda-129_arch-*, conda/environments/all_cuda-131_arch-*, conda/environments/clang_tidy_cuda-129_arch-x86_64.yaml, conda/environments/clang_tidy_cuda-131_arch-x86_64.yaml, conda/environments/cpp_all_cuda-129_arch-x86_64.yaml, conda/environments/cpp_all_cuda-131_arch-x86_64.yaml	Removed five CUDA development library packages (libcublas-dev, libcufft-dev, libcurand-dev, libcusolver-dev, libcusparse-dev) from each environment.
Conda Build Recipe conda/recipes/libnvforest/recipe.yaml	Removed CUDA library package dependencies (libcublas, libcufft, libcurand, libcusolver, libcusparse and their dev variants) from host/build and run dependencies across multiple output sections.
CMake Build Configuration cpp/CMakeLists.txt	Removed CUDA_STATIC_MATH_LIBRARIES-based static suffix setup and simplified target_compile_definitions by removing DISABLE_CUSPARSE_DEPRECATED PUBLIC flag.
Python Build System python/libnvforest/CMakeLists.txt	Removed CUDA runtime build-time path logic, rpath calculations based on CMAKE_CUDA_COMPILER_VERSION, and NVIDIA nccl lib path references.
Dependency Management dependencies.yaml	Restructured CUDA dependency handling by removing cuda_wheels from py_run includes, consolidating cuda and cuda_wheels blocks into a new cuda_version-driven matrix with explicit version mappings (12.2, 12.5, 12.8, 12.9, 13.0, 13.1) and separate cuda.common block for runtime components.
Python Project Configuration python/libnvforest/pyproject.toml, python/nvforest/pyproject.toml	Removed cuda-toolkit>=12,<14 dependency constraint from [project].dependencies.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~30 minutes

Possibly related PRs

• avoid packaging lib_treelite.{a,so} #77 — Updates CMake-based packaging configuration for python/libnvforest alongside similar build system changes.

Suggested reviewers

• hcho3

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately summarizes the three main changes: building wheels with CUDA 13.0.x, testing against multiple CTK versions, and dropping CUDA math library dependencies.
Description check	✅ Passed	The description clearly explains the rationale and details for each change, linking to relevant GitHub issues and justifying why dependencies are being removed.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

📝 Coding Plan

• Generate coding plan for human review comments

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[gforsyth]

/merge