Merge branch 'branch-25.05' into branch-25.08-merge-25.05

Author: rgsl888prabhu

.github/workflows/build.yaml

@@ -47,6 +47,15 @@ jobs:
       date: ${{ inputs.date }}
       sha: ${{ inputs.sha }}
       script: ci/build_python.sh
+  upload-conda:
+    needs: [cpp-build, python-build]
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/[email protected]
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      date: ${{ inputs.date }}
+      sha: ${{ inputs.sha }}
   wheel-build-cuopt-mps-parser:
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/[email protected]
@@ -60,6 +69,17 @@ jobs:
       package-name: cuopt_mps_parser
       package-type: python
       append-cuda-suffix: false
+  wheel-publish-cuopt-mps-parser:
+    needs: wheel-build-cuopt-mps-parser
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/[email protected]
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: cuopt_mps_parser
+      package-type: python
   wheel-build-libcuopt:
     needs: wheel-build-cuopt-mps-parser
     secrets: inherit
@@ -73,6 +93,17 @@ jobs:
       package-name: libcuopt
       package-type: cpp
       matrix_filter: map(select((.CUDA_VER | startswith("12")) and .PY_VER == "3.12"))
+  wheel-publish-libcuopt:
+    needs: wheel-build-libcuopt
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/[email protected]
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: libcuopt
+      package-type: cpp
   wheel-build-cuopt:
     needs: [wheel-build-cuopt-mps-parser, wheel-build-libcuopt]
     secrets: inherit
@@ -86,6 +117,17 @@ jobs:
       script: ci/build_wheel_cuopt.sh
       package-name: cuopt
       package-type: python
+  wheel-publish-cuopt:
+    needs: wheel-build-cuopt
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/[email protected]
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: cuopt
+      package-type: python
   wheel-build-cuopt-server:
     needs: wheel-build-cuopt
     secrets: inherit
@@ -99,6 +141,17 @@ jobs:
       script: ci/build_wheel_cuopt_server.sh
       package-name: cuopt_server
       package-type: python
+  wheel-publish-cuopt-server:
+    needs: wheel-build-cuopt-server
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/[email protected]
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: cuopt_server
+      package-type: python
   #docs-build:
   #  if: inputs.build_type == 'nightly' || github.ref_type == 'branch'
   #  needs: [python-build]
@@ -127,6 +180,17 @@ jobs:
       package-name: cuopt_sh_client
       package-type: python
       append-cuda-suffix: false
+  wheel-publish-cuopt-sh-client:
+    needs: wheel-build-cuopt-sh-client
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/[email protected]
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: cuopt_sh_client
+      package-type: python
   service-container:
     if: inputs.build_type == 'nightly'
     needs: [wheel-build-cuopt, wheel-build-cuopt-server]

.github/workflows/nightly.yaml

@@ -23,11 +23,15 @@ jobs:
           export DATE=$(date +%F)
           export SHA=$(gh api -q '.commit.sha' "repos/nvidia/cuopt/branches/${CUOPT_BRANCH}")
 
-          gh workflow run build.yaml \
+          RUN_ID=$(gh workflow run build.yaml \
             -f branch=${CUOPT_BRANCH} \
             -f sha=${SHA} \
             -f date=${DATE} \
-            -f build_type=nightly
+            -f build_type=nightly \
+            --json databaseId --jq '.databaseId')
+
+          # Wait for workflow to complete
+          gh run watch $RUN_ID
 
   trigger-test:
     runs-on: ubuntu-latest

README.md

@@ -63,6 +63,8 @@ For CUDA 12.x:
 pip install --extra-index-url=https://pypi.nvidia.com cuopt-server-cu12==25.5.* cuopt-sh-client==25.5.* nvidia-cuda-runtime-cu12==12.8.*
 ```
 
+Development wheels are available as nightlies, please update `--extra-index-url` to `https://pypi.anaconda.org/rapidsai-wheels-nightly/simple/` to install latest nightly packages.
+
 ### Conda
 
 cuOpt can be installed with conda (via [miniforge](https://github.com/conda-forge/miniforge)) from the `nvidia` channel:
@@ -74,19 +76,22 @@ Users who are used to conda env based workflows would benefit with conda package
 For CUDA 12.x:
 ```bash
 conda install -c rapidsai -c conda-forge -c nvidia \
-    cuopt-server=25.05 cuopt-sh-client=25.05 python=3.12 cuda-version=12.8
+    cuopt-server=25.05.* cuopt-sh-client=25.05.* python=3.12 cuda-version=12.8
 ```
 
 We also provide [nightly Conda packages](https://anaconda.org/rapidsai-nightly) built from the HEAD
-of our latest development branch.
+of our latest development branch. Just replace `-c rapidsai` with `-c rapidsai-nightly`.
 
 ### Container 
 
 Users can pull the cuOpt container from the NVIDIA container registry.
 
 ```bash
-docker pull nvidia/cuopt:25.5.0-cuda12.8-py312 
+docker pull nvidia/cuopt:latest-cuda12.8-py312 
 ```
+
+Note: The ``latest`` tag is the latest stable release of cuOpt. If you want to use a specific version, you can use the ``<version>-cuda12.8-py312`` tag. For example, to use cuOpt 25.5.0, you can use the ``25.5.0-cuda12.8-py312`` tag. Please refer to `cuOpt dockerhub page <https://hub.docker.com/r/nvidia/cuopt>`_ for the list of available tags.
+
 More information about the cuOpt container can be found [here](https://docs.nvidia.com/cuopt/user-guide/latest/cuopt-server/quick-start.html#container-from-docker-hub).
 
 Users who are using cuOpt for quick testing or research can use the cuOpt container. Alternatively, users who are planning to plug cuOpt as a service in their workflow can quickly start with the cuOpt container. But users are required to build security layers around the service to safeguard the service from untrusted users.

cpp/include/cuopt/linear_programming/constants.h

@@ -57,8 +57,9 @@
 #define CUOPT_MIP_RELATIVE_GAP            "mip_relative_gap"
 #define CUOPT_MIP_HEURISTICS_ONLY         "mip_heuristics_only"
 #define CUOPT_MIP_SCALING                 "mip_scaling"
-#define CUOPT_SOL_FILE                    "solution_file"
+#define CUOPT_SOLUTION_FILE               "solution_file"
 #define CUOPT_NUM_CPU_THREADS             "num_cpu_threads"
+#define CUOPT_USER_PROBLEM_FILE           "user_problem_file"
 
 /* @brief LP/MIP termination status constants */
 #define CUOPT_TERIMINATION_STATUS_NO_TERMINATION   0

cpp/include/cuopt/linear_programming/mip/solver_settings.hpp

@@ -92,6 +92,7 @@ class mip_solver_settings_t {
   bool log_to_console  = true;
   std::string log_file;
   std::string sol_file;
+  std::string user_problem_file;
 
   /** Initial primal solution */
   std::shared_ptr<rmm::device_uvector<f_t>> initial_solution_;

cpp/include/cuopt/linear_programming/pdlp/solver_settings.hpp

@@ -200,6 +200,7 @@ class pdlp_solver_settings_t {
   bool log_to_console{true};
   std::string log_file{""};
   std::string sol_file{""};
+  std::string user_problem_file{""};
   bool per_constraint_residual{false};
   bool crossover{false};
   bool save_best_primal_so_far{false};

cpp/include/cuopt/linear_programming/utilities/cython_solve.hpp

@@ -103,7 +103,8 @@ struct solver_ret_t {
 // Wrapper for solve to expose the API to cython.
 
 std::unique_ptr<solver_ret_t> call_solve(cuopt::mps_parser::data_model_view_t<int, double>*,
-                                         linear_programming::solver_settings_t<int, double>*);
+                                         linear_programming::solver_settings_t<int, double>*,
+                                         unsigned int flags = cudaStreamNonBlocking);
 
 std::pair<std::vector<std::unique_ptr<solver_ret_t>>, double> call_batch_solve(
   std::vector<cuopt::mps_parser::data_model_view_t<int, double>*>,

cpp/src/dual_simplex/branch_and_bound.cpp

@@ -463,8 +463,10 @@ mip_status_t branch_and_bound_t<i_t, f_t>::solve(mip_solution_t<i_t, f_t>& solut
     global_variables::mutex_upper.unlock();
     // We should be done here
     uncrush_primal_solution(original_problem, original_lp, incumbent.x, solution.x);
-    solution.objective   = incumbent.objective;
-    solution.lower_bound = lower_bound;
+    solution.objective          = incumbent.objective;
+    solution.lower_bound        = lower_bound;
+    solution.nodes_explored     = 0;
+    solution.simplex_iterations = root_relax_soln.iterations;
     settings.log.printf("Optimal solution found at root node. Objective %.16e. Time %.2f.\n",
                         compute_user_objective(original_lp, root_objective),
                         toc(start_time));

cpp/src/linear_programming/solve.cu

@@ -569,6 +569,11 @@ optimization_problem_solution_t<i_t, f_t> solve_lp(optimization_problem_t<i_t, f
                    problem.presolve_data.objective_offset,
                    problem.presolve_data.objective_scaling_factor);
 
+    if (settings.user_problem_file != "") {
+      CUOPT_LOG_INFO("Writing user problem to file: %s", settings.user_problem_file.c_str());
+      problem.write_as_mps(settings.user_problem_file);
+    }
+
     // Set the hyper-parameters based on the solver_settings
     if (use_pdlp_solver_mode) { set_pdlp_solver_mode(settings); }
 

cpp/src/linear_programming/utilities/cython_solve.cu

@@ -208,12 +208,13 @@ mip_ret_t call_solve_mip(
 
 std::unique_ptr<solver_ret_t> call_solve(
   cuopt::mps_parser::data_model_view_t<int, double>* data_model,
-  cuopt::linear_programming::solver_settings_t<int, double>* solver_settings)
+  cuopt::linear_programming::solver_settings_t<int, double>* solver_settings,
+  unsigned int flags)
 {
   raft::common::nvtx::range fun_scope("Call Solve");
 
   cudaStream_t stream;
-  RAFT_CUDA_TRY(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
+  RAFT_CUDA_TRY(cudaStreamCreateWithFlags(&stream, flags));
   const raft::handle_t handle_{stream};
 
   auto op_problem = data_model_to_optimization_problem(data_model, solver_settings, &handle_);
@@ -283,9 +284,11 @@ std::pair<std::vector<std::unique_ptr<solver_ret_t>>, double> call_batch_solve(
     solver_settings->set_parameter(CUOPT_METHOD, CUOPT_METHOD_PDLP);
   }
 
+  // Use a default stream instead of a non-blocking to avoid invalid operations while some CUDA
+  // Graph might be capturing in another stream
 #pragma omp parallel for num_threads(max_thread)
   for (std::size_t i = 0; i < size; ++i)
-    list[i] = std::move(call_solve(data_models[i], solver_settings));
+    list[i] = std::move(call_solve(data_models[i], solver_settings, cudaStreamDefault));
 
   auto end      = std::chrono::high_resolution_clock::now();
   auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(end - start_solver);