From ecdf895632bb64a00356f882353a5eaae981409f Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Tue, 28 Apr 2026 15:06:30 -0500
Subject: [PATCH 01/12] Use `check_array` in `cuml.fil`

---
 python/cuml/cuml/fil/fil.pyx | 47 +++++++++++++++++++++---------------
 1 file changed, 27 insertions(+), 20 deletions(-)

diff --git a/python/cuml/cuml/fil/fil.pyx b/python/cuml/cuml/fil/fil.pyx
index ce2f0371eb..1497d69cc9 100644
--- a/python/cuml/cuml/fil/fil.pyx
+++ b/python/cuml/cuml/fil/fil.pyx
@@ -6,19 +6,21 @@ import itertools
 import pathlib
 from time import perf_counter
 
+import cupy as cp
 import numpy as np
 import treelite.sklearn
+from cuda.bindings import runtime
 
 import cuml.internals.nvtx as nvtx
 from cuml.internals.array import CumlArray
 from cuml.internals.base import Base, get_handle
 from cuml.internals.device_type import DeviceType, DeviceTypeError
 from cuml.internals.global_settings import GlobalSettings
-from cuml.internals.input_utils import input_to_cuml_array
 from cuml.internals.mem_type import MemoryType
 from cuml.internals.mixins import CMajorInputTagMixin
 from cuml.internals.outputs import reflect
 from cuml.internals.treelite import safe_treelite_call
+from cuml.internals.validation import check_array
 
 from libc.stdint cimport uint32_t, uintptr_t
 from libcpp cimport bool
@@ -41,8 +43,6 @@ from cuml.internals.treelite cimport (
     TreeliteModelHandle,
 )
 
-from cuda.bindings import runtime
-
 
 cdef extern from "cuml/fil/forest_model.hpp" namespace "ML::fil" nogil:
     cdef cppclass forest_model:
@@ -124,12 +124,15 @@ def get_fil_device_type() -> DeviceType:
 
 cdef raft_proto_device_t get_fil_raft_proto_device_type(arr):
     """Get the current FIL device type as a raft_proto_device_t"""
-    cdef raft_proto_device_t dev
-    if arr.mem_type is MemoryType.device:
-        dev = raft_proto_device_t.gpu
+    if isinstance(arr, cp.ndarray):
+        return raft_proto_device_t.gpu
+    elif isinstance(arr, np.ndarray):
+        return raft_proto_device_t.cpu
     else:
-        dev = raft_proto_device_t.cpu
-    return dev
+        if arr.mem_type is MemoryType.device:
+            return raft_proto_device_t.gpu
+        else:
+            return raft_proto_device_t.cpu
 
 
 cdef class ForestInference_impl():
@@ -249,18 +252,22 @@ cdef class ForestInference_impl():
 
     def _predict(self, X, *, predict_type="default", preds=None, chunk_size=None):
         model_dtype = self.get_dtype()
+        mem_type = GlobalSettings().fil_memory_type
 
-        cdef uintptr_t in_ptr
-        in_arr, n_rows, _, _ = input_to_cuml_array(
+        X, index = check_array(
             X,
-            order='C',
-            convert_to_dtype=model_dtype,
-            convert_to_mem_type=GlobalSettings().fil_memory_type,
-            check_dtype=model_dtype
+            dtype=model_dtype,
+            order="C",
+            mem_type=mem_type.name,
+            return_index=True,
+            input_name="X",
+        )
+        n_rows = X.shape[0]
+
+        cdef raft_proto_device_t in_dev = get_fil_raft_proto_device_type(X)
+        cdef uintptr_t in_ptr = (
+            X.data.ptr if isinstance(X, cp.ndarray) else X.ctypes.data
         )
-        cdef raft_proto_device_t in_dev
-        in_dev = get_fil_raft_proto_device_type(in_arr)
-        in_ptr = in_arr.ptr
 
         cdef uintptr_t out_ptr
         cdef infer_kind infer_type_enum
@@ -283,14 +290,14 @@ cdef class ForestInference_impl():
                 output_shape,
                 model_dtype,
                 order='C',
-                index=in_arr.index,
-                mem_type=GlobalSettings().fil_memory_type,
+                index=index,
+                mem_type=mem_type,
             )
         else:
             # TODO(wphicks): Handle incorrect dtype/device/layout in C++
             if preds.shape != output_shape:
                 raise ValueError(f"If supplied, preds argument must have shape {output_shape}")
-            preds.index = in_arr.index
+            preds.index = index
         cdef raft_proto_device_t out_dev
         out_dev = get_fil_raft_proto_device_type(preds)
         out_ptr = preds.ptr

From a57e41347e8d79de8141b00de6b929b3003e6f0f Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Tue, 28 Apr 2026 15:07:37 -0500
Subject: [PATCH 02/12] Use `check_inputs` in `cuml.ensemble`

---
 .../cuml/ensemble/randomforest_common.pyx     |  4 +--
 .../cuml/ensemble/randomforestclassifier.py   | 28 ++++++++-----------
 .../cuml/ensemble/randomforestregressor.py    | 27 ++++++------------
 3 files changed, 23 insertions(+), 36 deletions(-)

diff --git a/python/cuml/cuml/ensemble/randomforest_common.pyx b/python/cuml/cuml/ensemble/randomforest_common.pyx
index 921608252e..ca5a741840 100644
--- a/python/cuml/cuml/ensemble/randomforest_common.pyx
+++ b/python/cuml/cuml/ensemble/randomforest_common.pyx
@@ -414,8 +414,8 @@ class BaseRandomForestModel(Base, InteropMixin):
         cdef bool is_classifier = self._estimator_type == "classifier"
         cdef bool is_float32 = X.dtype == np.float32
 
-        cdef uintptr_t X_ptr = X.ptr
-        cdef uintptr_t y_ptr = y.ptr
+        cdef uintptr_t X_ptr = X.data.ptr
+        cdef uintptr_t y_ptr = y.data.ptr
         cdef int n_rows = X.shape[0]
         cdef int n_cols = X.shape[1]
         cdef level_enum verbose = <level_enum> self._verbose_level
diff --git a/python/cuml/cuml/ensemble/randomforestclassifier.py b/python/cuml/cuml/ensemble/randomforestclassifier.py
index 3ec5bc7405..104f187b69 100644
--- a/python/cuml/cuml/ensemble/randomforestclassifier.py
+++ b/python/cuml/cuml/ensemble/randomforestclassifier.py
@@ -1,9 +1,6 @@
 # SPDX-FileCopyrightText: Copyright (c) 2019-2026, NVIDIA CORPORATION.
 # SPDX-License-Identifier: Apache-2.0
 
-import cupy as cp
-import numpy as np
-
 import cuml.internals
 import cuml.internals.nvtx as nvtx
 from cuml.common.array_descriptor import CumlArrayDescriptor
@@ -11,10 +8,9 @@
 from cuml.common.doc_utils import generate_docstring, insert_into_docstring
 from cuml.ensemble.randomforest_common import BaseRandomForestModel
 from cuml.internals.array import CumlArray
-from cuml.internals.input_utils import input_to_cuml_array
 from cuml.internals.interop import UnsupportedOnGPU
 from cuml.internals.mixins import ClassifierMixin
-from cuml.internals.validation import check_features, check_y
+from cuml.internals.validation import check_features, check_inputs
 from cuml.metrics import accuracy_score
 
 
@@ -225,19 +221,20 @@ def fit(self, X, y, *, convert_dtype=True) -> "RandomForestClassifier":
             y to be of dtype int32. This will increase memory used for
             the method.
         """
-        y, classes = check_y(y, dtype=cp.int32, return_classes=True)
-        X_m = input_to_cuml_array(
+        X, y, classes = check_inputs(
+            self,
             X,
-            convert_to_dtype=(np.float32 if convert_dtype else None),
-            check_dtype=[np.float32, np.float64],
+            y,
+            dtype=("float32", "float64"),
+            convert_dtype=convert_dtype,
             order="F",
-            check_rows=y.shape[0],
-        ).array
+            y_dtype="int32",
+            return_classes=True,
+            reset=True,
+        )
         self.classes_ = classes
         self.n_classes_ = len(classes)
-        y_m = CumlArray(data=y)
-
-        return self._fit_forest(X_m, y_m)
+        return self._fit_forest(X, y)
 
     @nvtx.annotate(
         message="predict RF-Classifier @randomforestclassifier.pyx",
@@ -309,8 +306,7 @@ def predict_proba(
         align_bytes=None,
     ) -> CumlArray:
         """
-        Predicts class probabilities for X. This function uses the GPU
-        implementation of predict.
+        Predicts class probabilities for X.
 
         Parameters
         ----------
diff --git a/python/cuml/cuml/ensemble/randomforestregressor.py b/python/cuml/cuml/ensemble/randomforestregressor.py
index d232a9f633..ef31f4c216 100644
--- a/python/cuml/cuml/ensemble/randomforestregressor.py
+++ b/python/cuml/cuml/ensemble/randomforestregressor.py
@@ -1,16 +1,13 @@
 # SPDX-FileCopyrightText: Copyright (c) 2019-2026, NVIDIA CORPORATION.
 # SPDX-License-Identifier: Apache-2.0
-import numpy as np
-
 import cuml.internals.nvtx as nvtx
-from cuml.common import input_to_cuml_array
 from cuml.common.array_descriptor import CumlArrayDescriptor
 from cuml.common.doc_utils import generate_docstring, insert_into_docstring
 from cuml.ensemble.randomforest_common import BaseRandomForestModel
 from cuml.internals.array import CumlArray
 from cuml.internals.mixins import RegressorMixin
 from cuml.internals.outputs import reflect, run_in_internal_context
-from cuml.internals.validation import check_features
+from cuml.internals.validation import check_features, check_inputs
 from cuml.metrics import r2_score
 
 
@@ -187,22 +184,16 @@ def fit(self, X, y, *, convert_dtype=True) -> "RandomForestRegressor":
         Perform Random Forest Regression on the input data
 
         """
-        X_m = input_to_cuml_array(
+        X, y = check_inputs(
+            self,
             X,
-            convert_to_dtype=(np.float32 if convert_dtype else None),
-            check_dtype=[np.float32, np.float64],
-            order="F",
-        ).array
-
-        y_m = input_to_cuml_array(
             y,
-            convert_to_dtype=(X_m.dtype if convert_dtype else None),
-            check_dtype=X_m.dtype,
-            check_rows=X_m.shape[0],
-            check_cols=1,
-        ).array
-
-        return self._fit_forest(X_m, y_m)
+            dtype=("float32", "float64"),
+            convert_dtype=convert_dtype,
+            order="F",
+            reset=True,
+        )
+        return self._fit_forest(X, y)
 
     @nvtx.annotate(
         message="predict RF-Regressor @randomforestclassifier.pyx",

From 971e268df410f93c382cf0a2f4864d6916b0237c Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Tue, 28 Apr 2026 15:20:38 -0500
Subject: [PATCH 03/12] Update sklearn compatibility xfails

---
 .../cuml/tests/test_sklearn_compatibility.py  | 27 +++++--------------
 1 file changed, 6 insertions(+), 21 deletions(-)

diff --git a/python/cuml/tests/test_sklearn_compatibility.py b/python/cuml/tests/test_sklearn_compatibility.py
index ed0fc0b84f..398e1d24f3 100644
--- a/python/cuml/tests/test_sklearn_compatibility.py
+++ b/python/cuml/tests/test_sklearn_compatibility.py
@@ -118,17 +118,14 @@
     RandomForestRegressor: {
         "check_estimator_tags_renamed": "No support for modern tags infrastructure",
         "check_do_not_raise_errors_in_init_or_set_params": "RandomForestRegressor raises errors in init or set_params",
-        "check_dtype_object": "RandomForestRegressor does not handle object dtype",
-        "check_estimators_empty_data_messages": "RandomForestRegressor does not handle empty data",
-        "check_estimators_nan_inf": "RandomForestRegressor does not check for NaN and inf",
-        "check_regressors_train": "RandomForestRegressor does not handle list inputs",
-        "check_regressors_train(readonly_memmap=True)": "RandomForestRegressor does not handle readonly memmap",
-        "check_regressors_train(readonly_memmap=True,X_dtype=float32)": "RandomForestRegressor does not handle readonly memmap with float32",
         "check_regressor_data_not_an_array": "RandomForestRegressor does not handle non-array data",
-        "check_supervised_y_2d": "RandomForestRegressor does not handle 2D y",
-        "check_supervised_y_no_nan": "RandomForestRegressor does not check for NaN in y",
         "check_dict_unchanged": "RandomForestRegressor modifies input dictionaries",
-        "check_requires_y_none": "RandomForestRegressor does not handle y=None",
+    },
+    RandomForestClassifier: {
+        "check_estimator_tags_renamed": "No support for modern tags infrastructure",
+        "check_do_not_raise_errors_in_init_or_set_params": "RandomForestClassifier raises errors in init or set_params",
+        "check_classifier_data_not_an_array": "RandomForestClassifier does not handle non-array data",
+        "check_dict_unchanged": "RandomForestClassifier modifies input dictionaries",
     },
     KNeighborsClassifier: {
         "check_estimator_tags_renamed": "No support for modern tags infrastructure",
@@ -139,18 +136,6 @@
         "check_classifier_data_not_an_array": "KNeighborsClassifier does not handle non-array data",
         "check_classifiers_train": "KNeighborsClassifier does not validate input data properly",
     },
-    RandomForestClassifier: {
-        "check_estimator_tags_renamed": "No support for modern tags infrastructure",
-        "check_do_not_raise_errors_in_init_or_set_params": "RandomForestClassifier raises errors in init or set_params",
-        "check_dtype_object": "RandomForestClassifier does not handle object dtype",
-        "check_estimators_empty_data_messages": "RandomForestClassifier does not handle empty data",
-        "check_estimators_nan_inf": "RandomForestClassifier does not check for NaN and inf",
-        "check_classifier_data_not_an_array": "RandomForestClassifier does not handle non-array data",
-        "check_classifiers_train": "RandomForestClassifier does not handle list inputs",
-        "check_classifiers_train(readonly_memmap=True)": "RandomForestClassifier does not handle readonly memmap",
-        "check_classifiers_train(readonly_memmap=True,X_dtype=float32)": "RandomForestClassifier does not handle readonly memmap with float32",
-        "check_dict_unchanged": "RandomForestClassifier modifies input dictionaries",
-    },
     KNeighborsRegressor: {
         "check_estimator_tags_renamed": "No support for modern tags infrastructure",
         "check_do_not_raise_errors_in_init_or_set_params": "KNeighborsRegressor raises errors in init or set_params",

From 9b373253b74a672afca68d326d06a4febaf16356 Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Tue, 28 Apr 2026 15:13:19 -0500
Subject: [PATCH 04/12] Add `RandomForestClassifier.predict_log_proba`

---
 .../cuml/ensemble/randomforestclassifier.py   | 52 +++++++++++++++++++
 python/cuml/tests/test_random_forest.py       |  9 ++++
 2 files changed, 61 insertions(+)

diff --git a/python/cuml/cuml/ensemble/randomforestclassifier.py b/python/cuml/cuml/ensemble/randomforestclassifier.py
index 104f187b69..4634915692 100644
--- a/python/cuml/cuml/ensemble/randomforestclassifier.py
+++ b/python/cuml/cuml/ensemble/randomforestclassifier.py
@@ -1,5 +1,6 @@
 # SPDX-FileCopyrightText: Copyright (c) 2019-2026, NVIDIA CORPORATION.
 # SPDX-License-Identifier: Apache-2.0
+import cupy as cp
 
 import cuml.internals
 import cuml.internals.nvtx as nvtx
@@ -339,6 +340,57 @@ def predict_proba(
         check_features(self, X)
         return fil.predict_proba(X)
 
+    @insert_into_docstring(
+        parameters=[("dense", "(n_samples, n_features)")],
+        return_values=[("dense", "(n_samples, 1)")],
+    )
+    @cuml.internals.reflect
+    def predict_log_proba(
+        self,
+        X,
+        *,
+        convert_dtype=True,
+        layout="depth_first",
+        default_chunk_size=None,
+        align_bytes=None,
+    ) -> CumlArray:
+        """
+        Predicts log class probabilities for X.
+
+        Parameters
+        ----------
+        X : {}
+        convert_dtype : bool (default = True)
+            When True, automatically convert the input to the data type used
+            to train the model. This may increase memory usage.
+        layout : string (default = 'depth_first')
+            Specifies the in-memory layout of nodes in FIL forests. Options:
+            'depth_first', 'layered', 'breadth_first'.
+        default_chunk_size : int, optional (default = None)
+            Determines how batches are further subdivided for parallel processing.
+            The optimal value depends on hardware, model, and batch size.
+            If None, will be automatically determined.
+        align_bytes : int, optional (default = None)
+            If specified, trees will be padded such that their in-memory size is
+            a multiple of this value. This can improve performance by guaranteeing
+            that memory reads from trees begin on a cache line boundary.
+            Typical values are 0 or 128 on GPU and 0 or 64 on CPU.
+
+        Returns
+        -------
+        y : {}
+        """
+        preds = self.predict_proba(
+            X,
+            convert_dtype=convert_dtype,
+            layout=layout,
+            default_chunk_size=default_chunk_size,
+            align_bytes=align_bytes,
+        )
+        out = preds.to_output("cupy")
+        cp.log(out, out=out)
+        return CumlArray(data=out, index=preds.index)
+
     @nvtx.annotate(
         message="score RF-Classifier @randomforestclassifier.pyx",
         domain="cuml_python",
diff --git a/python/cuml/tests/test_random_forest.py b/python/cuml/tests/test_random_forest.py
index 2c9b68a98e..5c1e9c043e 100644
--- a/python/cuml/tests/test_random_forest.py
+++ b/python/cuml/tests/test_random_forest.py
@@ -984,6 +984,15 @@ def test_rf_multiclass_classifier_gtil_integration(tmpdir):
     np.testing.assert_almost_equal(out_prob, expected_prob, decimal=5)
 
 
+def test_classifier_predict_log_proba():
+    X, y = make_classification(random_state=42)
+    model = curfc(random_state=42).fit(X, y)
+    proba = model.predict_proba(X)
+    sol = np.log(proba)
+    log_proba = model.predict_log_proba(X)
+    np.testing.assert_allclose(log_proba, sol, rtol=1e-5)
+
+
 @pytest.mark.parametrize(
     "estimator, make_data",
     [

From bb440d8b73887b393b4bc18cf049a0e015c72010 Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Tue, 28 Apr 2026 15:20:16 -0500
Subject: [PATCH 05/12] Cleanup overrides for `sklearn.ensemble`

---
 .../cuml/accel/_overrides/sklearn/ensemble.py | 83 +++++++++++--------
 1 file changed, 47 insertions(+), 36 deletions(-)

diff --git a/python/cuml/cuml/accel/_overrides/sklearn/ensemble.py b/python/cuml/cuml/accel/_overrides/sklearn/ensemble.py
index a2c84d54c7..72dcb353ee 100644
--- a/python/cuml/cuml/accel/_overrides/sklearn/ensemble.py
+++ b/python/cuml/cuml/accel/_overrides/sklearn/ensemble.py
@@ -1,39 +1,67 @@
 #
-# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION.
 # SPDX-License-Identifier: Apache-2.0
 #
 
 import cuml.ensemble
 from cuml.accel.estimator_proxy import ProxyBase
-from cuml.internals.input_utils import input_to_cuml_array
 from cuml.internals.interop import UnsupportedOnGPU
+from cuml.internals.validation import check_array
 
 __all__ = ("RandomForestRegressor", "RandomForestClassifier")
 
 
-class RandomForestRegressor(ProxyBase):
-    _gpu_class = cuml.ensemble.RandomForestRegressor
+class _RandomForestMixin:
+    def _check_inputs(self, X, y=None, sample_weight=None):
+        # Fallback to CPU if NaN in X
+        try:
+            check_array(
+                X, mem_type=None, order=None, ensure_2d=False, input_name="X"
+            )
+        except ValueError as exc:
+            if "NaN" in str(exc):
+                raise UnsupportedOnGPU(
+                    "Missing values are not supported"
+                ) from None
+            raise
 
-    def _gpu_fit(self, X, y, sample_weight=None):
         if sample_weight is not None:
             raise UnsupportedOnGPU("`sample_weight` is not supported")
 
-        try:
-            y = input_to_cuml_array(y, convert_to_mem_type=False)[0]
-        except ValueError:
-            raise
-        else:
+        if y is not None:
+            y = check_array(
+                y,
+                mem_type=None,
+                order=None,
+                ensure_2d=False,
+                ensure_all_finite=False,
+                input_name="y",
+            )
             if len(y.shape) > 1 and y.shape[1] > 1:
+                if isinstance(self, RandomForestClassifier) and self.oob_score:
+                    raise ValueError(
+                        "The type of target cannot be used to compute OOB estimates"
+                    )
                 raise UnsupportedOnGPU(
                     "Multi-output targets are not supported"
                 )
+
+    def _gpu_fit(self, X, y, sample_weight=None):
+        self._check_inputs(X, y, sample_weight=sample_weight)
         return self._gpu.fit(X, y)
 
+    def _gpu_predict(self, X):
+        self._check_inputs(X)
+        return self._gpu.predict(X)
+
     def _gpu_score(self, X, y, sample_weight=None):
-        if sample_weight is not None:
-            raise UnsupportedOnGPU("`sample_weight` is not supported")
+        self._check_inputs(X, y, sample_weight=sample_weight)
         return self._gpu.score(X, y)
 
+
+class RandomForestRegressor(ProxyBase, _RandomForestMixin):
+    _gpu_class = cuml.ensemble.RandomForestRegressor
+
     def __len__(self):
         return self._call_method("__len__")
 
@@ -44,33 +72,16 @@ def __getitem__(self, index):
         return self._call_method("__getitem__", index)
 
 
-class RandomForestClassifier(ProxyBase):
+class RandomForestClassifier(ProxyBase, _RandomForestMixin):
     _gpu_class = cuml.ensemble.RandomForestClassifier
 
-    def _gpu_fit(self, X, y, sample_weight=None):
-        if sample_weight is not None:
-            raise UnsupportedOnGPU("`sample_weight` is not supported")
+    def _gpu_predict_proba(self, X):
+        self._check_inputs(X)
+        return self._gpu.predict_proba(X)
 
-        try:
-            y = input_to_cuml_array(y, convert_to_mem_type=False)[0]
-        except ValueError:
-            raise
-        else:
-            if len(y.shape) > 1 and y.shape[1] > 1:
-                if self.oob_score:
-                    raise ValueError(
-                        "The type of target cannot be used to compute OOB estimates"
-                    )
-                else:
-                    raise UnsupportedOnGPU(
-                        "Multi-output targets are not supported"
-                    )
-        return self._gpu.fit(X, y)
-
-    def _gpu_score(self, X, y, sample_weight=None):
-        if sample_weight is not None:
-            raise UnsupportedOnGPU("`sample_weight` is not supported")
-        return self._gpu.score(X, y)
+    def _gpu_predict_log_proba(self, X):
+        self._check_inputs(X)
+        return self._gpu.predict_log_proba(X)
 
     def __len__(self):
         return self._call_method("__len__")

From 37595cc0cf4b7571fa47d0666b9f6b99429c21b5 Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Tue, 28 Apr 2026 15:19:41 -0500
Subject: [PATCH 06/12] Don't reload CPU fit models on GPU in cuml.accel

---
 python/cuml/cuml/accel/estimator_proxy.py     | 19 ++++++++++++-------
 .../cuml_accel_tests/test_estimator_proxy.py  | 17 +++++++++++++++++
 2 files changed, 29 insertions(+), 7 deletions(-)

diff --git a/python/cuml/cuml/accel/estimator_proxy.py b/python/cuml/cuml/accel/estimator_proxy.py
index 1bbf5e2c00..0a923390b5 100644
--- a/python/cuml/cuml/accel/estimator_proxy.py
+++ b/python/cuml/cuml/accel/estimator_proxy.py
@@ -60,7 +60,7 @@ def _reconstruct(self):
         # will always serialize it by value rather than by reference. This allows
         # the saved model to be loaded in an environment without cuml installed,
         # where it will load as the CPU model directly.
-        def reconstruct(cls_path, cpu_model):
+        def reconstruct(cls_path, cpu_model, load_on_gpu):
             """Reconstruct a serialized estimator.
 
             Returns a proxy estimator if `cuml.accel` is installed, falling back
@@ -82,7 +82,7 @@ def reconstruct(cls_path, cpu_model):
                 # Return the CPU estimator directly
                 return cpu_model
             # `cuml.accel` is installed, reconstruct a proxy estimator
-            return cls._reconstruct_from_cpu(cpu_model)
+            return cls._reconstruct_from_cpu(cpu_model, load_on_gpu)
 
         return reconstruct
 
@@ -241,7 +241,7 @@ def _sync_attrs_to_cpu(self) -> None:
             )
 
     @classmethod
-    def _reconstruct_from_cpu(cls, cpu):
+    def _reconstruct_from_cpu(cls, cpu, load_on_gpu=True):
         """Reconstruct a proxy estimator from its CPU counterpart.
 
         Primarily used when unpickling serialized proxy estimators."""
@@ -249,8 +249,9 @@ def _reconstruct_from_cpu(cls, cpu):
         self = cls.__new__(cls)
         self._cpu = cpu
         self._synced = False
-        if is_fitted(self._cpu):
-            # This is a fit estimator. Try to convert model back to GPU
+        if load_on_gpu and is_fitted(self._cpu):
+            # This estimator is fit and should be loaded on GPU. Try to convert
+            # model back to GPU.
             try:
                 self._gpu = self._gpu_class.from_sklearn(self._cpu)
             except UnsupportedOnGPU:
@@ -259,7 +260,7 @@ def _reconstruct_from_cpu(cls, cpu):
                 # Supported on GPU, clear fit attributes from CPU to release host memory
                 self._cpu = sklearn.clone(self._cpu)
         else:
-            # Estimator is unfit, delay GPU init until needed
+            # Estimator is unfit or should remain on CPU
             self._gpu = None
         return self
 
@@ -427,7 +428,11 @@ def __reduce__(self):
         self._sync_attrs_to_cpu()
         return (
             _reconstruct_proxy,
-            (self._gpu_class._cpu_class_path, self._cpu),
+            (
+                self._gpu_class._cpu_class_path,
+                self._cpu,
+                self._gpu is not None,
+            ),
         )
 
     def __getattr__(self, name: str) -> Any:
diff --git a/python/cuml/cuml_accel_tests/test_estimator_proxy.py b/python/cuml/cuml_accel_tests/test_estimator_proxy.py
index 0effc9819f..5579bcd6c5 100644
--- a/python/cuml/cuml_accel_tests/test_estimator_proxy.py
+++ b/python/cuml/cuml_accel_tests/test_estimator_proxy.py
@@ -407,6 +407,23 @@ def test_pickle_cpu_fit():
     assert not model2._synced
 
 
+def test_pickle_cpu_fit_but_params_supported():
+    """Test that a model fit on CPU isn't reloaded to the GPU on unpickle, even
+    if the hyperparameters are supported."""
+    X, y = make_regression(n_samples=10)
+    X[0, 0] = np.nan
+    model = RandomForestRegressor().fit(X, y)
+    # ensure the test case is one that isn't GPU accelerated
+    assert model._gpu is None
+
+    model2 = pickle.loads(pickle.dumps(model))
+    # GPU model doesn't exist
+    assert model2._gpu is None
+    # CPU model has fit attributes
+    assert model2._cpu.n_features_in_
+    assert not model2._synced
+
+
 def test_unpickle_cuml_accel_not_active():
     """Unpickling in an process without cuml.accel enabled uses the CPU model"""
     X, y = make_classification(n_samples=10)

From 8f98440d421ac3d1da002b4f0b23d5cb48eba43c Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Tue, 28 Apr 2026 15:20:53 -0500
Subject: [PATCH 07/12] Update xfail list

---
 .../upstream/scikit-learn/xfail-list.yaml       | 17 -----------------
 1 file changed, 17 deletions(-)

diff --git a/python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml b/python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml
index bf3f648ec9..8d9778285e 100644
--- a/python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml
+++ b/python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml
@@ -207,13 +207,9 @@
   - "sklearn.ensemble.tests.test_forest::test_min_samples_leaf[RandomForestRegressor]"
   - "sklearn.ensemble.tests.test_forest::test_min_samples_split[RandomForestClassifier]"
   - "sklearn.ensemble.tests.test_forest::test_min_samples_split[RandomForestRegressor]"
-  - "sklearn.ensemble.tests.test_forest::test_missing_value_is_predictive[RandomForestClassifier]"
-  - "sklearn.ensemble.tests.test_forest::test_missing_value_is_predictive[RandomForestRegressor]"
-  - "sklearn.ensemble.tests.test_forest::test_missing_values_is_resilient[make_regression-RandomForestRegressor]"
   - "sklearn.ensemble.tests.test_forest::test_oob_not_computed_twice[RandomForestClassifier]"
   - "sklearn.ensemble.tests.test_forest::test_oob_not_computed_twice[RandomForestRegressor]"
   - "sklearn.ensemble.tests.test_forest::test_poisson_y_positive_check"
-  - "sklearn.ensemble.tests.test_forest::test_probability[RandomForestClassifier]"
   - "sklearn.ensemble.tests.test_forest::test_warm_start[RandomForestClassifier]"
   - "sklearn.ensemble.tests.test_forest::test_warm_start[RandomForestRegressor]"
   - "sklearn.ensemble.tests.test_forest::test_warm_start_oob[RandomForestRegressor]"
@@ -914,11 +910,6 @@
   - "sklearn.neighbors.tests.test_neighbors::test_neighbors_metrics[42-float64-canberra]"
   - "sklearn.neighbors.tests.test_neighbors::test_neighbors_metrics[42-float64-haversine]"
   - "sklearn.neighbors.tests.test_neighbors::test_neighbors_metrics[42-float64-minkowski]"
-- reason: ValueError when input data is non-standard type
-  marker: cuml_accel_rf_input_validation_non_standard_types
-  tests:
-  - "sklearn.tests.test_common::test_estimators[RandomForestClassifier()-check_requires_y_none]"
-  - "sklearn.tests.test_common::test_estimators[RandomForestRegressor()-check_requires_y_none]"
 - reason: Multi-output targets not supported
   marker: cuml_accel_rf_multioutput_targets_not_supported
   tests:
@@ -1016,16 +1007,8 @@
   - "sklearn.tests.test_common::test_estimators[NearestNeighbors()-check_estimators_nan_inf]"
   - "sklearn.tests.test_common::test_estimators[PCA()-check_fit2d_1feature]"
   - "sklearn.tests.test_common::test_estimators[PCA()-check_fit2d_1sample]"
-  - "sklearn.tests.test_common::test_estimators[RandomForestClassifier()-check_classifier_data_not_an_array]"
   - "sklearn.tests.test_common::test_estimators[RandomForestClassifier()-check_classifiers_multilabel_output_format_decision_function]"
-  - "sklearn.tests.test_common::test_estimators[RandomForestClassifier()-check_classifiers_train(readonly_memmap=True)]"
-  - "sklearn.tests.test_common::test_estimators[RandomForestClassifier()-check_classifiers_train]"
-  - "sklearn.tests.test_common::test_estimators[RandomForestClassifier()-check_dtype_object]"
-  - "sklearn.tests.test_common::test_estimators[RandomForestClassifier()-check_estimators_empty_data_messages]"
-  - "sklearn.tests.test_common::test_estimators[RandomForestRegressor()-check_dtype_object]"
-  - "sklearn.tests.test_common::test_estimators[RandomForestRegressor()-check_estimators_empty_data_messages]"
   - "sklearn.tests.test_common::test_estimators[RandomForestRegressor()-check_regressor_data_not_an_array]"
-  - "sklearn.tests.test_common::test_estimators[RandomForestRegressor()-check_supervised_y_no_nan]"
   - "sklearn.tests.test_common::test_estimators[SpectralEmbedding()-check_dtype_object]"
   - "sklearn.tests.test_common::test_estimators[SpectralEmbedding()-check_estimators_nan_inf]"
   - "sklearn.tests.test_common::test_estimators[TSNE()-check_dtype_object]"

From 9855936dad3484a035597171cfc617db7a329012 Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Tue, 28 Apr 2026 15:21:02 -0500
Subject: [PATCH 08/12] Update ensemble limitations

---
 docs/source/cuml-accel/limitations.rst | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/docs/source/cuml-accel/limitations.rst b/docs/source/cuml-accel/limitations.rst
index a41ba733df..e2c6152278 100644
--- a/docs/source/cuml-accel/limitations.rst
+++ b/docs/source/cuml-accel/limitations.rst
@@ -197,6 +197,7 @@ RandomForestClassifier
 - If ``class_weight`` is not ``None``.
 - If ``sample_weight`` is passed to ``fit`` or ``score``.
 - If ``X`` is sparse.
+- If ``X`` contains missing values (represented as ``NaN``).
 - If ``y`` is a multi-output target.
 
 RandomForestRegressor
@@ -213,6 +214,7 @@ RandomForestRegressor
 - If ``ccp_alpha`` is not ``0``.
 - If ``sample_weight`` is passed to ``fit`` or ``score``.
 - If ``X`` is sparse.
+- If ``X`` contains missing values (represented as ``NaN``).
 - If ``y`` is a multi-output target.
 
 

From 137be01c8ba70209af736a0874570da4f7a49259 Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Tue, 28 Apr 2026 16:22:53 -0500
Subject: [PATCH 09/12] Fixup estimator proxy pickle change

---
 python/cuml/cuml/accel/estimator_proxy.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/python/cuml/cuml/accel/estimator_proxy.py b/python/cuml/cuml/accel/estimator_proxy.py
index 0a923390b5..4c4f8b40a4 100644
--- a/python/cuml/cuml/accel/estimator_proxy.py
+++ b/python/cuml/cuml/accel/estimator_proxy.py
@@ -96,8 +96,8 @@ def __reduce__(self):
 
         return (pickle.loads, (cloudpickle.dumps(self._reconstruct),))
 
-    def __call__(self, cls_path, cpu):
-        return self._reconstruct(cls_path, cpu)
+    def __call__(self, cls_path, cpu, load_on_gpu):
+        return self._reconstruct(cls_path, cpu, load_on_gpu)
 
 
 _reconstruct_proxy = _ReconstructProxy()

From 7e67063d195592e966226b7b89e1ae678fc04ee8 Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Wed, 29 Apr 2026 11:03:28 -0500
Subject: [PATCH 10/12] Only disallow NaN in cuml.ensemble

---
 python/cuml/cuml/ensemble/randomforest_common.pyx |  1 +
 python/cuml/cuml/fil/fil.pyx                      | 13 ++++++++++++-
 2 files changed, 13 insertions(+), 1 deletion(-)

diff --git a/python/cuml/cuml/ensemble/randomforest_common.pyx b/python/cuml/cuml/ensemble/randomforest_common.pyx
index ca5a741840..9f2e213531 100644
--- a/python/cuml/cuml/ensemble/randomforest_common.pyx
+++ b/python/cuml/cuml/ensemble/randomforest_common.pyx
@@ -408,6 +408,7 @@ class BaseRandomForestModel(Base, InteropMixin):
             layout=layout,
             default_chunk_size=default_chunk_size,
             align_bytes=align_bytes,
+            allow_nan=False,
         )
 
     def _fit_forest(self, X, y):
diff --git a/python/cuml/cuml/fil/fil.pyx b/python/cuml/cuml/fil/fil.pyx
index 1497d69cc9..fab537df52 100644
--- a/python/cuml/cuml/fil/fil.pyx
+++ b/python/cuml/cuml/fil/fil.pyx
@@ -139,6 +139,7 @@ cdef class ForestInference_impl():
     cdef forest_model model
     cdef raft_proto_handle_t raft_proto_handle
     cdef object raft_handle
+    cdef bool allow_nan
 
     def __cinit__(
         self,
@@ -149,6 +150,7 @@ cdef class ForestInference_impl():
         use_double_precision=None,
         mem_type=None,
         device_id=None,
+        allow_nan=True,
     ):
         # Store reference to RAFT handle to control lifetime, since raft_proto
         # handle keeps a pointer to it
@@ -156,6 +158,7 @@ cdef class ForestInference_impl():
         self.raft_proto_handle = raft_proto_handle_t(
             <raft_handle_t*><size_t>self.raft_handle.getHandle()
         )
+        self.allow_nan = allow_nan
         if mem_type is None:
             mem_type = GlobalSettings().fil_memory_type
         else:
@@ -260,6 +263,7 @@ cdef class ForestInference_impl():
             order="C",
             mem_type=mem_type.name,
             return_index=True,
+            ensure_all_finite="allow-nan" if self.allow_nan else True,
             input_name="X",
         )
         n_rows = X.shape[0]
@@ -460,6 +464,9 @@ class ForestInference(Base, CMajorInputTagMixin):
         For GPU execution, the device on which to load and execute this
         model. If set to None, use the currently active device.
         For CPU execution, this value is currently ignored.
+    allow_nan : bool, default=True
+        Whether to allow NaN values in X during inference. If False, an error
+        will be raised in the presence of NaN inputs.
     """
 
     def _reload_model(self):
@@ -614,6 +621,7 @@ class ForestInference(Base, CMajorInputTagMixin):
         align_bytes=None,
         precision='single',
         device_id=None,
+        allow_nan=True,
     ):
         super().__init__(verbose=verbose, output_type=output_type)
         self.is_classifier = is_classifier
@@ -623,6 +631,7 @@ class ForestInference(Base, CMajorInputTagMixin):
         self.precision = precision
         self.device_id = device_id
         self.treelite_model = treelite_model
+        self.allow_nan = allow_nan
         self._load_to_fil(device_id=self.device_id)
 
     def _load_to_fil(self, mem_type=None, device_id=None):
@@ -657,7 +666,8 @@ class ForestInference(Base, CMajorInputTagMixin):
                 align_bytes=self.align_bytes,
                 use_double_precision=self._use_double_precision_,
                 mem_type=mem_type,
-                device_id=self.device_id
+                device_id=self.device_id,
+                allow_nan=self.allow_nan
             )
 
             if mem_type is MemoryType.device:
@@ -1356,4 +1366,5 @@ class ForestInference(Base, CMajorInputTagMixin):
             "align_bytes",
             "precision",
             "device_id",
+            "allow_nan",
         ]

From 1e69ba8556b0c22544ced469cae96f8ab910ccb1 Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Wed, 29 Apr 2026 11:23:31 -0500
Subject: [PATCH 11/12] Respond to feedback

---
 python/cuml/cuml/fil/fil.pyx | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/python/cuml/cuml/fil/fil.pyx b/python/cuml/cuml/fil/fil.pyx
index fab537df52..6f2aeb66df 100644
--- a/python/cuml/cuml/fil/fil.pyx
+++ b/python/cuml/cuml/fil/fil.pyx
@@ -329,8 +329,8 @@ cdef class ForestInference_impl():
                 <double*> out_ptr,
                 <double*> in_ptr,
                 n_rows,
-                in_dev,
                 out_dev,
+                in_dev,
                 infer_type_enum,
                 chunk_specification
             )
@@ -630,8 +630,8 @@ class ForestInference(Base, CMajorInputTagMixin):
         self.layout = layout
         self.precision = precision
         self.device_id = device_id
-        self.treelite_model = treelite_model
         self.allow_nan = allow_nan
+        self.treelite_model = treelite_model
         self._load_to_fil(device_id=self.device_id)
 
     def _load_to_fil(self, mem_type=None, device_id=None):

From 2bd239531aad6010389cd95bfb61651d7542e0a6 Mon Sep 17 00:00:00 2001
From: Jim Crist-Harif <jcristharif@gmail.com>
Date: Wed, 29 Apr 2026 18:41:18 -0500
Subject: [PATCH 12/12] FIL allows infinity by default

---
 .../cuml/ensemble/randomforest_common.pyx     |  2 +-
 python/cuml/cuml/fil/fil.pyx                  | 23 ++++++++++---------
 2 files changed, 13 insertions(+), 12 deletions(-)

diff --git a/python/cuml/cuml/ensemble/randomforest_common.pyx b/python/cuml/cuml/ensemble/randomforest_common.pyx
index 9f2e213531..a886aeb4b7 100644
--- a/python/cuml/cuml/ensemble/randomforest_common.pyx
+++ b/python/cuml/cuml/ensemble/randomforest_common.pyx
@@ -408,7 +408,7 @@ class BaseRandomForestModel(Base, InteropMixin):
             layout=layout,
             default_chunk_size=default_chunk_size,
             align_bytes=align_bytes,
-            allow_nan=False,
+            ensure_all_finite=True,
         )
 
     def _fit_forest(self, X, y):
diff --git a/python/cuml/cuml/fil/fil.pyx b/python/cuml/cuml/fil/fil.pyx
index 6f2aeb66df..446443dd74 100644
--- a/python/cuml/cuml/fil/fil.pyx
+++ b/python/cuml/cuml/fil/fil.pyx
@@ -139,7 +139,7 @@ cdef class ForestInference_impl():
     cdef forest_model model
     cdef raft_proto_handle_t raft_proto_handle
     cdef object raft_handle
-    cdef bool allow_nan
+    cdef object ensure_all_finite
 
     def __cinit__(
         self,
@@ -150,7 +150,7 @@ cdef class ForestInference_impl():
         use_double_precision=None,
         mem_type=None,
         device_id=None,
-        allow_nan=True,
+        ensure_all_finite=False,
     ):
         # Store reference to RAFT handle to control lifetime, since raft_proto
         # handle keeps a pointer to it
@@ -158,7 +158,7 @@ cdef class ForestInference_impl():
         self.raft_proto_handle = raft_proto_handle_t(
             <raft_handle_t*><size_t>self.raft_handle.getHandle()
         )
-        self.allow_nan = allow_nan
+        self.ensure_all_finite = ensure_all_finite
         if mem_type is None:
             mem_type = GlobalSettings().fil_memory_type
         else:
@@ -263,7 +263,7 @@ cdef class ForestInference_impl():
             order="C",
             mem_type=mem_type.name,
             return_index=True,
-            ensure_all_finite="allow-nan" if self.allow_nan else True,
+            ensure_all_finite=self.ensure_all_finite,
             input_name="X",
         )
         n_rows = X.shape[0]
@@ -464,9 +464,10 @@ class ForestInference(Base, CMajorInputTagMixin):
         For GPU execution, the device on which to load and execute this
         model. If set to None, use the currently active device.
         For CPU execution, this value is currently ignored.
-    allow_nan : bool, default=True
-        Whether to allow NaN values in X during inference. If False, an error
-        will be raised in the presence of NaN inputs.
+    ensure_all_finite : bool or 'allow-nan', default=False
+        If True, an error will be raised if non-finite values are found in the
+        input. If 'allow-nan', an error will be raised if infinite values are
+        found (but not for NaN). If False then ``check_all_finite`` is skipped.
     """
 
     def _reload_model(self):
@@ -621,7 +622,7 @@ class ForestInference(Base, CMajorInputTagMixin):
         align_bytes=None,
         precision='single',
         device_id=None,
-        allow_nan=True,
+        ensure_all_finite=False,
     ):
         super().__init__(verbose=verbose, output_type=output_type)
         self.is_classifier = is_classifier
@@ -630,7 +631,7 @@ class ForestInference(Base, CMajorInputTagMixin):
         self.layout = layout
         self.precision = precision
         self.device_id = device_id
-        self.allow_nan = allow_nan
+        self.ensure_all_finite = ensure_all_finite
         self.treelite_model = treelite_model
         self._load_to_fil(device_id=self.device_id)
 
@@ -667,7 +668,7 @@ class ForestInference(Base, CMajorInputTagMixin):
                 use_double_precision=self._use_double_precision_,
                 mem_type=mem_type,
                 device_id=self.device_id,
-                allow_nan=self.allow_nan
+                ensure_all_finite=self.ensure_all_finite
             )
 
             if mem_type is MemoryType.device:
@@ -1366,5 +1367,5 @@ class ForestInference(Base, CMajorInputTagMixin):
             "align_bytes",
             "precision",
             "device_id",
-            "allow_nan",
+            "ensure_all_finite",
         ]