Remove the deprecated convert_dtype kwarg

python/cuml/cuml/cluster/agglomerative.pyx

@@ -136,7 +136,7 @@ class AgglomerativeClustering(ClusterMixin, CMajorInputTagMixin, Base):
 
     @generate_docstring()
     @mlfunc(set_input_type=True)
-    def fit(self, X, y=None, *, convert_dtype="deprecated") -> "AgglomerativeClustering":
+    def fit(self, X, y=None) -> "AgglomerativeClustering":
         """
         Fit the hierarchical clustering from features.
         """
@@ -146,7 +146,6 @@ class AgglomerativeClustering(ClusterMixin, CMajorInputTagMixin, Base):
             X,
             order="C",
             dtype="float32",
-            convert_dtype=convert_dtype,
             ensure_min_samples=2,
             reset=True,
         )

python/cuml/cuml/cluster/dbscan.pyx

@@ -305,7 +305,6 @@ class DBSCAN(InteropMixin,
         sample_weight=None,
         *,
         out_dtype="int32",
-        convert_dtype="deprecated"
     ) -> "DBSCAN":
         """
         Perform DBSCAN clustering from features.
@@ -327,7 +326,6 @@ class DBSCAN(InteropMixin,
             X,
             sample_weight=sample_weight,
             dtype=("float32", "float64"),
-            convert_dtype=convert_dtype,
             order="C",
             return_index=True,
             reset=True,
@@ -482,7 +480,6 @@ class DBSCAN(InteropMixin,
         sample_weight=None,
         *,
         out_dtype="int32",
-        convert_dtype="deprecated",
     ):
         """
         Performs clustering on X and returns cluster labels.
@@ -499,10 +496,5 @@ class DBSCAN(InteropMixin,
             negative weight may inhibit its eps-neighbor from being core.
             default: None (which is equivalent to weight 1 for all samples).
         """
-        self.fit(
-            X,
-            sample_weight=sample_weight,
-            out_dtype=out_dtype,
-            convert_dtype=convert_dtype
-        )
+        self.fit(X, sample_weight=sample_weight, out_dtype=out_dtype)
         return self.labels_

python/cuml/cuml/cluster/hdbscan/hdbscan.pyx

@@ -903,7 +903,7 @@ class HDBSCAN(InteropMixin, ClusterMixin, CMajorInputTagMixin, Base):
 
     @generate_docstring()
     @mlfunc(set_input_type=True)
-    def fit(self, X, y=None, *, convert_dtype="deprecated") -> "HDBSCAN":
+    def fit(self, X, y=None) -> "HDBSCAN":
         """
         Fit HDBSCAN model from features.
         """
@@ -923,7 +923,6 @@ class HDBSCAN(InteropMixin, ClusterMixin, CMajorInputTagMixin, Base):
             self,
             X,
             dtype="float32",
-            convert_dtype=convert_dtype,
             mem_type=mem_type,
             ensure_min_samples=2,
             return_index=True,
@@ -1177,12 +1176,7 @@ def all_points_membership_vectors(clusterer, int batch_size=4096):
 
 
 @mlfunc(model_arg="clusterer", array_arg="points_to_predict", preserve_index=True)
-def membership_vector(
-    clusterer,
-    points_to_predict,
-    int batch_size=4096,
-    convert_dtype="deprecated",
-):
+def membership_vector(clusterer, points_to_predict, int batch_size=4096):
     """
     Predict soft cluster membership. The result produces a vector
     for each point in ``points_to_predict`` that gives a probability that
@@ -1221,7 +1215,6 @@ def membership_vector(
         clusterer,
         points_to_predict,
         dtype="float32",
-        convert_dtype=convert_dtype,
         order="C",
     )
     cdef int n_prediction_points = points_to_predict.shape[0]
@@ -1261,7 +1254,7 @@ def membership_vector(
 
 
 @mlfunc(model_arg="clusterer", array_arg="points_to_predict", preserve_index=True)
-def approximate_predict(clusterer, points_to_predict, convert_dtype="deprecated"):
+def approximate_predict(clusterer, points_to_predict):
     """Predict the cluster label of new points. The returned labels
     will be those of the original clustering found by ``clusterer``,
     and therefore are not (necessarily) the cluster labels that would
@@ -1305,7 +1298,6 @@ def approximate_predict(clusterer, points_to_predict, convert_dtype="deprecated"
         clusterer,
         points_to_predict,
         dtype="float32",
-        convert_dtype=convert_dtype,
         order="C",
     )
     cdef int n_prediction_points = points_to_predict.shape[0]

python/cuml/cuml/cluster/kmeans.pyx

@@ -724,7 +724,7 @@ class KMeans(InteropMixin,
 
     @generate_docstring()
     @mlfunc(set_input_type=True)
-    def fit(self, X, y=None, sample_weight=None, *, convert_dtype="deprecated") -> "KMeans":
+    def fit(self, X, y=None, sample_weight=None) -> "KMeans":
         """
         Compute k-means clustering with X.
 
@@ -745,7 +745,6 @@ class KMeans(InteropMixin,
             X,
             sample_weight=sample_weight,
             dtype=("float32", "float64"),
-            convert_dtype=convert_dtype,
             order=None,
             mem_type=None,
             reset=True,
@@ -783,7 +782,6 @@ class KMeans(InteropMixin,
                 self.init,
                 order="C",
                 dtype=X.dtype,
-                convert_dtype=convert_dtype,
             ).copy()
             if centers.shape[0] != self.n_clusters:
                 raise ValueError(
@@ -1006,7 +1004,7 @@ class KMeans(InteropMixin,
         """
         return self.fit(X, sample_weight=sample_weight).labels_
 
-    def _predict_labels_inertia(self, X, convert_dtype="deprecated", sample_weight=None):
+    def _predict_labels_inertia(self, X, sample_weight=None):
         """
         Predict the closest cluster each sample in X belongs to.
 
@@ -1017,13 +1015,6 @@ class KMeans(InteropMixin,
             Acceptable formats: cuDF DataFrame, NumPy ndarray, Numba device
             ndarray, cuda array interface compliant array like CuPy
 
-        convert_dtype : bool, default="deprecated"
-            .. deprecated:: 26.08
-                `convert_dtype` was deprecated in version 26.08 and will be
-                removed in version 26.10. cuML only copies input arrays when
-                necessary (e.g. to unify dtypes), there is no reason to provide
-                this keyword going forward.
-
         sample_weight : array-like (device or host) shape = (n_samples,), default=None # noqa
             The weights for each observation in X. If None, all observations
             are assigned equal weight.
@@ -1042,7 +1033,6 @@ class KMeans(InteropMixin,
             X,
             sample_weight=sample_weight,
             dtype=self.cluster_centers_.dtype,
-            convert_dtype=convert_dtype,
             order="C",
         )
         if sample_weight is None:
@@ -1068,20 +1058,20 @@ class KMeans(InteropMixin,
                                        'description': 'Cluster indexes',
                                        'shape': '(n_samples, 1)'})
     @mlfunc(preserve_index=True)
-    def predict(self, X, *, convert_dtype="deprecated"):
+    def predict(self, X):
         """
         Predict the closest cluster each sample in X belongs to.
 
         """
-        labels, _ = self._predict_labels_inertia(X, convert_dtype=convert_dtype)
+        labels, _ = self._predict_labels_inertia(X)
         return labels
 
     @generate_docstring(return_values={'name': 'X_new',
                                        'type': 'dense',
                                        'description': 'Transformed data',
                                        'shape': '(n_samples, n_clusters)'})
     @mlfunc(preserve_index=True)
-    def transform(self, X, *, convert_dtype="deprecated"):
+    def transform(self, X):
         """
         Transform X to a cluster-distance space.
 
@@ -1091,7 +1081,6 @@ class KMeans(InteropMixin,
             self,
             X,
             dtype=self.cluster_centers_.dtype,
-            convert_dtype=convert_dtype,
             order="C",
         )
 
@@ -1178,28 +1167,24 @@ class KMeans(InteropMixin,
                                                         of X on the K-means \
                                                         objective.'})
     @mlfunc(convert_output=False)
-    def score(self, X, y=None, sample_weight=None, *, convert_dtype="deprecated"):
+    def score(self, X, y=None, sample_weight=None):
         """
         Opposite of the value of X on the K-means objective.
 
         """
 
-        inertia = self._predict_labels_inertia(
-            X, convert_dtype=convert_dtype, sample_weight=sample_weight
-        )[1]
+        inertia = self._predict_labels_inertia(X, sample_weight=sample_weight)[1]
         return -1 * inertia
 
     @generate_docstring(return_values={'name': 'X_new',
                                        'type': 'dense',
                                        'description': 'Transformed data',
                                        'shape': '(n_samples, n_clusters)'})
     @mlfunc(preserve_index=True)
-    def fit_transform(
-        self, X, y=None, sample_weight=None, *, convert_dtype="deprecated"
-    ):
+    def fit_transform(self, X, y=None, sample_weight=None):
         """
         Compute clustering and transform X to cluster-distance space.
 
         """
         self.fit(X, sample_weight=sample_weight)
-        return self.transform(X, convert_dtype=convert_dtype)
+        return self.transform(X)

python/cuml/cuml/common/doc_utils.py

@@ -1,5 +1,5 @@
 #
-# SPDX-FileCopyrightText: Copyright (c) 2019-2026, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2019-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 #
 
@@ -54,12 +54,6 @@
     "    Acceptable dense formats: CUDA array interface compliant objects like\n"  # noqa
     "    CuPy, cuDF DataFrame/Series, NumPy ndarray and Pandas\n"
     "    DataFrame/Series.",
-    "convert_dtype": "convert_dtype : bool, optional (default = 'deprecated')\n"
-    "    .. deprecated:: 26.08\n"
-    "        `convert_dtype` was deprecated in version 26.08 and will be removed\n"
-    "        in version 26.10. cuML only copies input arrays when necessary\n"
-    "        (e.g. to unify dtypes), there is no reason to provide this keyword\n"
-    "        going forward.\n",
     "sample_weight": "sample_weight : array-like (device or host) shape = (n_samples,), default={default}\n"  # noqa
     "    The weights for each observation in X. If None, all observations\n"
     "    are assigned equal weight.\n"
@@ -107,7 +101,6 @@
 _return_values_possible_values = ["name", "type", "shape", "description"]
 
 _simple_params = [
-    "convert_dtype",
     "return_sparse",
     "sparse_tol",
     "sample_weight",
@@ -134,7 +127,6 @@ def generate_docstring(
     Currently auto detected variables include:
     - X
     - y
-    - convert_dtype
     - sample_weights
     - return_sparse
     - sparse_tol

python/cuml/cuml/covariance/empirical_covariance.py

@@ -147,9 +147,7 @@ def __init__(
         self.assume_centered = assume_centered
 
     @mlfunc(set_input_type=True)
-    def fit(
-        self, X, y=None, *, convert_dtype="deprecated"
-    ) -> "EmpiricalCovariance":
+    def fit(self, X, y=None) -> "EmpiricalCovariance":
         """Fit the maximum likelihood covariance estimator to X.
 
         Parameters
@@ -159,12 +157,6 @@ def fit(
             and `n_features` is the number of features.
         y : Ignored
             Not used, present for API consistency.
-        convert_dtype : bool, default="deprecated"
-            .. deprecated:: 26.08
-                `convert_dtype` was deprecated in version 26.08 and will be
-                removed in version 26.10. cuML only copies input arrays when
-                necessary (e.g. to unify dtypes), there is no reason to provide
-                this keyword going forward.
 
         Returns
         -------
@@ -175,7 +167,6 @@ def fit(
             self,
             X,
             dtype=("float32", "float64"),
-            convert_dtype=convert_dtype,
             reset=True,
         )
         if X.shape[0] == 1:

python/cuml/cuml/covariance/ledoit_wolf.py

@@ -228,7 +228,7 @@ def __init__(
         self.block_size = block_size
 
     @mlfunc(set_input_type=True)
-    def fit(self, X, y=None, *, convert_dtype="deprecated") -> "LedoitWolf":
+    def fit(self, X, y=None) -> "LedoitWolf":
         """Fit the Ledoit-Wolf shrunk covariance model to X.
 
         Parameters
@@ -238,12 +238,6 @@ def fit(self, X, y=None, *, convert_dtype="deprecated") -> "LedoitWolf":
             and `n_features` is the number of features.
         y : Ignored
             Not used, present for API consistency.
-        convert_dtype : bool, default="deprecated"
-            .. deprecated:: 26.08
-                `convert_dtype` was deprecated in version 26.08 and will be
-                removed in version 26.10. cuML only copies input arrays when
-                necessary (e.g. to unify dtypes), there is no reason to provide
-                this keyword going forward.
 
         Returns
         -------
@@ -254,7 +248,6 @@ def fit(self, X, y=None, *, convert_dtype="deprecated") -> "LedoitWolf":
             self,
             X,
             dtype=("float32", "float64"),
-            convert_dtype=convert_dtype,
             reset=True,
         )
         if X.shape[0] == 1:

python/cuml/cuml/dask/ensemble/base.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2021-2026, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2021-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 #
 
@@ -88,7 +88,7 @@ def _estimators_per_worker(self, n_estimators):
             n_estimators_per_worker[i] = n_estimators_per_worker[i] + 1
         return n_estimators_per_worker
 
-    def _fit(self, model, dataset, convert_dtype, broadcast_data):
+    def _fit(self, model, dataset, broadcast_data):
         data = DistributedDataHandler.create(dataset, client=self.client)
         self.active_workers = data.workers
         self.datatype = data.datatype
@@ -114,7 +114,6 @@ def _fit(self, model, dataset, convert_dtype, broadcast_data):
                     _func_fit,
                     model[worker],
                     combined_data if broadcast_data else worker_data,
-                    convert_dtype,
                     workers=[worker],
                     pure=False,
                 )
@@ -305,10 +304,10 @@ def apply_reduction(self, reduce, partial_infs, datatype, delayed):
             return delayed_res.persist()
 
 
-def _func_fit(model, input_data, convert_dtype):
+def _func_fit(model, input_data):
     X = concatenate([item[0] for item in input_data])
     y = concatenate([item[1] for item in input_data])
-    return model.fit(X, y, convert_dtype=convert_dtype)
+    return model.fit(X, y)
 
 
 def _func_predict_partial(model, input_data, **kwargs):