Default CPU benchmark estimators to max parallelism

conda/recipes/cuml/recipe.yaml

@@ -101,6 +101,7 @@ requirements:
     - numba >=0.60.0,<0.65.0
     - numba-cuda >=0.22.2,<0.29.0
     - numpy >=1.23,<3.0
+    - pyyaml
     - scikit-learn >=1.4
     - scipy >=1.14.0
     - packaging

python/cuml/cuml/benchmark/algorithms.py

@@ -3,6 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 #
 import functools
+import inspect
 import shutil
 import sys
 import tempfile
@@ -55,6 +56,22 @@
 cuml_metrics = None
 treelite = None
 
+
+def _with_default_cpu_parallelism(cpu_class, args):
+    """Use max CPU parallelism when the estimator exposes n_jobs."""
+    if cpu_class is None:
+        return args
+    try:
+        signature = inspect.signature(cpu_class)
+    except (TypeError, ValueError):
+        return args
+
+    updated_args = dict(args)
+    if "n_jobs" in signature.parameters and "n_jobs" not in updated_args:
+        updated_args["n_jobs"] = -1
+    return updated_args
+
+
 # GPU-specific helper functions (only available when GPU libs present)
 _build_cpu_skl_classifier = None
 _build_fil_classifier = None
@@ -224,6 +241,7 @@ def run_cpu(self, data, bench_args={}, **override_setup_args):
 
         all_args = {**self.shared_args, **self.cpu_args}
         all_args = {**all_args, **override_setup_args}
+        all_args = _with_default_cpu_parallelism(self.cpu_class, all_args)
 
         if "cpu_setup_result" not in all_args:
             cpu_obj = self.cpu_class(**all_args)
@@ -262,6 +280,7 @@ def run_cuml(self, data, bench_args={}, **override_setup_args):
     def setup_cpu(self, data, **override_args):
         all_args = {**self.shared_args, **self.cpu_args}
         all_args = {**all_args, **override_args}
+        all_args = _with_default_cpu_parallelism(self.cpu_class, all_args)
         if self.setup_cpu_func is not None:
             return {
                 "cpu_setup_result": self.setup_cpu_func(
@@ -670,7 +689,7 @@ def all_algorithms():
                 HDBSCAN,
                 cuml_HDBSCAN,
                 shared_args={},
-                cpu_args={},
+                cpu_args={"core_dist_n_jobs": -1},
                 name="HDBSCAN",
                 accepts_labels=False,
             )

python/cuml/cuml/benchmark/cli.py

@@ -79,6 +79,18 @@ def build_parser():
         "--quiet", "-q", action="store_false", dest="verbose", default=True
     )
     parser.add_argument("--csv", nargs="?")
+    parser.add_argument(
+        "--config",
+        type=str,
+        default=None,
+        help="Path to a YAML benchmark config file",
+    )
+    parser.add_argument(
+        "--profile",
+        type=str,
+        default=None,
+        help="Named profile to select from a YAML benchmark config",
+    )
     parser.add_argument("--dataset", default="blobs")
     parser.add_argument(
         "--skip-cpu",
@@ -88,6 +100,15 @@ def build_parser():
     parser.add_argument(
         "--skip-gpu", action="store_true", help="Skip GPU/cuML benchmarks"
     )
+    parser.add_argument(
+        "--backends",
+        type=str,
+        default=None,
+        help=(
+            "Comma-separated execution backends to run. Supported values: "
+            "cpu,gpu. Overrides manifest backends in config mode."
+        ),
+    )
     parser.add_argument("--input-type", default="numpy")
     parser.add_argument(
         "--test-split",