From 89b1fcf50875ae5eda712d01d16e1f7e6176f002 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Tue, 2 Jun 2026 01:27:01 +0000
Subject: [PATCH 01/14] refactor(cudf_polars): use cudf_streaming for
 cudf-specific streaming imports

---
 python/cudf_polars/cudf_polars/engine/core.py |  4 ++--
 python/cudf_polars/cudf_polars/engine/dask.py |  2 +-
 python/cudf_polars/cudf_polars/engine/ray.py  |  2 +-
 python/cudf_polars/cudf_polars/engine/spmd.py |  4 ++--
 .../actor_graph/collectives/allgather.py      |  4 ++--
 .../actor_graph/collectives/shuffle.py        | 20 +++++++++----------
 .../streaming/actor_graph/collectives/sort.py | 10 +++++-----
 .../cudf_polars/streaming/actor_graph/core.py |  4 ++--
 .../streaming/actor_graph/groupby.py          | 10 +++++-----
 .../cudf_polars/streaming/actor_graph/io.py   |  6 +++---
 .../cudf_polars/streaming/actor_graph/join.py | 20 +++++++++----------
 .../streaming/actor_graph/nodes.py            | 10 +++++-----
 .../cudf_polars/streaming/actor_graph/over.py | 10 +++++-----
 .../streaming/actor_graph/repartition.py      | 10 +++++-----
 .../streaming/actor_graph/union.py            |  4 ++--
 .../streaming/actor_graph/utils.py            | 12 +++++------
 python/cudf_polars/pyproject.toml             |  1 +
 17 files changed, 67 insertions(+), 66 deletions(-)

diff --git a/python/cudf_polars/cudf_polars/engine/core.py b/python/cudf_polars/cudf_polars/engine/core.py
index 79914d12fe8d..f7881b5ea95a 100644
--- a/python/cudf_polars/cudf_polars/engine/core.py
+++ b/python/cudf_polars/cudf_polars/engine/core.py
@@ -14,12 +14,12 @@
 from typing import TYPE_CHECKING, Any, ClassVar, Self, TypeVar
 
 import cuda.core
+from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.coll import AllGather
 from rapidsmpf.config import Options, get_environment_variables
 from rapidsmpf.memory.packed_data import PackedData
 from rapidsmpf.statistics import Statistics
 from rapidsmpf.streaming.core.actor import run_actor_network
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
 
 import polars as pl
 
@@ -42,10 +42,10 @@
     from concurrent.futures import Executor, ThreadPoolExecutor
 
     import rapidsmpf.config
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.memory.buffer_resource import BufferResource
     from rapidsmpf.streaming.core.context import Context
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
 
     from cudf_polars.dsl.ir import IR
     from cudf_polars.streaming.base import PartitionInfo
diff --git a/python/cudf_polars/cudf_polars/engine/dask.py b/python/cudf_polars/cudf_polars/engine/dask.py
index 014627b82c48..f37a5db27998 100644
--- a/python/cudf_polars/cudf_polars/engine/dask.py
+++ b/python/cudf_polars/cudf_polars/engine/dask.py
@@ -44,8 +44,8 @@
 if TYPE_CHECKING:
     from collections.abc import Callable
 
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from rapidsmpf.communicator.communicator import Communicator
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
 
     from cudf_polars.dsl.ir import IR
     from cudf_polars.engine.core import T
diff --git a/python/cudf_polars/cudf_polars/engine/ray.py b/python/cudf_polars/cudf_polars/engine/ray.py
index edf1905e916d..686767230e17 100644
--- a/python/cudf_polars/cudf_polars/engine/ray.py
+++ b/python/cudf_polars/cudf_polars/engine/ray.py
@@ -41,8 +41,8 @@
     import uuid
     from collections.abc import Callable
 
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from rapidsmpf.communicator.communicator import Communicator
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
     from ray.actor import ActorHandle
 
     from cudf_polars.dsl.ir import IR
diff --git a/python/cudf_polars/cudf_polars/engine/spmd.py b/python/cudf_polars/cudf_polars/engine/spmd.py
index d334e9f1c95c..c6f057d48b12 100644
--- a/python/cudf_polars/cudf_polars/engine/spmd.py
+++ b/python/cudf_polars/cudf_polars/engine/spmd.py
@@ -10,13 +10,13 @@
 from concurrent.futures import ThreadPoolExecutor
 from typing import TYPE_CHECKING, Any, cast
 
+from cudf_streaming.integrations.partition import unpack_and_concat
 from rapidsmpf import bootstrap
 from rapidsmpf.coll import AllGather
 from rapidsmpf.communicator.single import (
     new_communicator as single_communicator,
 )
 from rapidsmpf.communicator.ucxx import barrier
-from rapidsmpf.integrations.cudf.partition import unpack_and_concat
 from rapidsmpf.memory.packed_data import PackedData
 from rapidsmpf.progress_thread import ProgressThread
 from rapidsmpf.rmm_resource_adaptor import RmmResourceAdaptor
@@ -51,9 +51,9 @@
     import uuid
     from collections.abc import Callable
 
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.config import Options
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
 
     import polars as pl
 
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
index 911e3f270920..5e899958d95f 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
@@ -6,16 +6,16 @@
 
 from typing import TYPE_CHECKING
 
-from rapidsmpf.integrations.cudf.partition import unpack_and_concat
+from cudf_streaming.integrations.partition import unpack_and_concat
 from rapidsmpf.memory.packed_data import PackedData
 from rapidsmpf.streaming.coll.allgather import AllGather
 
 from pylibcudf.contiguous_split import pack
 
 if TYPE_CHECKING:
+    from cudf_streaming.streaming.table_chunk import TableChunk
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.streaming.core.context import Context
-    from rapidsmpf.streaming.cudf.table_chunk import TableChunk
 
     import pylibcudf as plc
     from rmm.pylibrmm.stream import Stream
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py
index 7da5259afd13..99c2f38d5069 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py
@@ -6,24 +6,24 @@
 
 from typing import TYPE_CHECKING, Any
 
-from rapidsmpf.communicator.single import new_communicator as single_comm
-from rapidsmpf.config import Options, get_environment_variables
-from rapidsmpf.integrations.cudf.partition import (
+from cudf_streaming.integrations.partition import (
     partition_and_pack as py_partition_and_pack,
     split_and_pack as py_split_and_pack,
     unpack_and_concat as py_unpack_and_concat,
 )
+from cudf_streaming.streaming.channel_metadata import (
+    ChannelMetadata,
+    HashScheme,
+    Partitioning,
+)
+from cudf_streaming.streaming.table_chunk import TableChunk
+from rapidsmpf.communicator.single import new_communicator as single_comm
+from rapidsmpf.config import Options, get_environment_variables
 from rapidsmpf.shuffler import PartitionAssignment
 from rapidsmpf.streaming.coll.shuffler import ShufflerAsync
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.context import Context
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import (
-    ChannelMetadata,
-    HashScheme,
-    Partitioning,
-)
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
 
 import pylibcudf as plc
 import pylibcudf.partitioning
@@ -445,7 +445,7 @@ async def shuffle_actor(
 
     This node combines partition_and_pack, shuffler, and unpack_and_concat
     into a single Python node using rapidsmpf.shuffler.Shuffler and utilities
-    from rapidsmpf.integrations.cudf.partition.
+    from cudf_streaming.integrations.partition.
 
     Parameters
     ----------
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py
index 75087141c9b2..cec4901d5641 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py
@@ -6,16 +6,16 @@
 
 from typing import TYPE_CHECKING
 
-from rapidsmpf.shuffler import PartitionAssignment
-from rapidsmpf.streaming.core.actor import define_actor
-from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import (
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     OrderKey,
     OrderScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
+from cudf_streaming.streaming.table_chunk import TableChunk
+from rapidsmpf.shuffler import PartitionAssignment
+from rapidsmpf.streaming.core.actor import define_actor
+from rapidsmpf.streaming.core.message import Message
 
 import polars as pl
 
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py
index fb7cc4e1439a..0160681262a1 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py
@@ -30,12 +30,12 @@
 if TYPE_CHECKING:
     from collections.abc import MutableMapping
 
+    from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+    from cudf_streaming.streaming.table_chunk import TableChunk
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.streaming.core.channel import Channel
     from rapidsmpf.streaming.core.context import Context
     from rapidsmpf.streaming.core.leaf_actor import DeferredMessages
-    from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-    from rapidsmpf.streaming.cudf.table_chunk import TableChunk
 
     import polars as pl
 
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py
index c9f65da127c1..932c3662b99b 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py
@@ -7,16 +7,16 @@
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any
 
+from cudf_streaming.streaming.channel_metadata import (
+    ChannelMetadata,
+    HashScheme,
+)
+from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.communicator.single import new_communicator as single_comm
 from rapidsmpf.config import Options, get_environment_variables
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.context import Context
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import (
-    ChannelMetadata,
-    HashScheme,
-)
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
 
 import pylibcudf as plc
 
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
index 7f4db5e38bb3..8abf0a25c9fa 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
@@ -8,13 +8,13 @@
 import math
 from typing import TYPE_CHECKING, Any
 
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.memory.memory_reservation import opaque_memory_usage
 from rapidsmpf.streaming.core.memory_reserve_or_wait import (
     reserve_memory,
 )
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
 
 import pylibcudf as plc
 
@@ -481,7 +481,7 @@ def make_rapidsmpf_read_parquet_node(
     The RapidsMPF read parquet node, or None if the predicate cannot be
     converted to a parquet filter (caller should fall back to scan_node).
     """
-    from rapidsmpf.streaming.cudf.parquet import Filter, read_parquet
+    from cudf_streaming.streaming.parquet import Filter, read_parquet
 
     # Build ParquetReaderOptions
     try:
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py
index 0130146576a6..8cac2a693bac 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py
@@ -7,22 +7,22 @@
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any, Literal
 
-from rapidsmpf.memory.memory_reservation import opaque_memory_usage
-from rapidsmpf.streaming.core.actor import define_actor
-from rapidsmpf.streaming.core.memory_reserve_or_wait import (
-    reserve_memory,
-)
-from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.bloom_filter import BloomFilter
-from rapidsmpf.streaming.cudf.channel_metadata import (
+from cudf_streaming.streaming.bloom_filter import BloomFilter
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import (
+from cudf_streaming.streaming.table_chunk import (
     TableChunk,
     make_table_chunks_available_or_wait,
 )
+from rapidsmpf.memory.memory_reservation import opaque_memory_usage
+from rapidsmpf.streaming.core.actor import define_actor
+from rapidsmpf.streaming.core.memory_reserve_or_wait import (
+    reserve_memory,
+)
+from rapidsmpf.streaming.core.message import Message
 
 import pylibcudf as plc
 from pylibcudf.hashing import LIBCUDF_DEFAULT_HASH_SEED
@@ -60,10 +60,10 @@
     from collections.abc import Iterable, MutableMapping
     from types import CoroutineType
 
+    from cudf_streaming.streaming.bloom_filter import BloomFilterChunk
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.streaming.core.channel import Channel
     from rapidsmpf.streaming.core.context import Context
-    from rapidsmpf.streaming.cudf.bloom_filter import BloomFilterChunk
 
     from cudf_polars.dsl.ir import IR, IRExecutionContext
     from cudf_polars.streaming.actor_graph.dispatch import SubNetGenerator
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/nodes.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/nodes.py
index 9432fd7295d8..777fb02c7a64 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/nodes.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/nodes.py
@@ -7,16 +7,16 @@
 import asyncio
 from typing import TYPE_CHECKING, Any, cast
 
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import (
+    TableChunk,
+    make_table_chunks_available_or_wait,
+)
 from rapidsmpf.memory.buffer import MemoryType
 from rapidsmpf.memory.memory_reservation import opaque_memory_usage
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.message import Message
 from rapidsmpf.streaming.core.spillable_messages import SpillableMessages
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import (
-    TableChunk,
-    make_table_chunks_available_or_wait,
-)
 
 from cudf_polars.containers import DataFrame
 from cudf_polars.dsl.ir import IR, Empty
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
index d1569d561b21..623e1c9c5cc1 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
@@ -40,15 +40,15 @@
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any, ClassVar, cast
 
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import (
+    TableChunk,
+    make_table_chunks_available_or_wait,
+)
 from rapidsmpf.memory.memory_reservation import opaque_memory_usage
 from rapidsmpf.shuffler import PartitionAssignment
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import (
-    TableChunk,
-    make_table_chunks_available_or_wait,
-)
 
 import polars as pl
 
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/repartition.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/repartition.py
index fcfc897e1728..ccd62b8c1c15 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/repartition.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/repartition.py
@@ -7,14 +7,14 @@
 import math
 from typing import TYPE_CHECKING, Any
 
-from rapidsmpf.memory.memory_reservation import opaque_memory_usage
-from rapidsmpf.streaming.core.actor import define_actor
-from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import (
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import (
     TableChunk,
     make_table_chunks_available_or_wait,
 )
+from rapidsmpf.memory.memory_reservation import opaque_memory_usage
+from rapidsmpf.streaming.core.actor import define_actor
+from rapidsmpf.streaming.core.message import Message
 
 from cudf_polars.containers import DataFrame
 from cudf_polars.streaming.actor_graph.collectives.allgather import AllGatherManager
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/union.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/union.py
index a2664fa3a798..bf14a23f1c15 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/union.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/union.py
@@ -6,9 +6,9 @@
 
 from typing import TYPE_CHECKING, Any
 
+from cudf_streaming.streaming.channel_metadata import ChannelMetadata
+from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import ChannelMetadata
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
 
 from cudf_polars.dsl.ir import Union
 from cudf_polars.streaming.actor_graph.dispatch import (
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py
index ac87b16d5872..1679391574ee 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py
@@ -16,21 +16,21 @@
 from functools import reduce
 from typing import TYPE_CHECKING, Any, Literal, TypeAlias, cast
 
-from rapidsmpf.memory.memory_reservation import opaque_memory_usage
-from rapidsmpf.memory.packed_data import PackedData
-from rapidsmpf.streaming.coll.allgather import AllGather
-from rapidsmpf.streaming.core.message import Message
-from rapidsmpf.streaming.cudf.channel_metadata import (
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
     OrderKey,
     OrderScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import (
+from cudf_streaming.streaming.table_chunk import (
     TableChunk,
     make_table_chunks_available_or_wait,
 )
+from rapidsmpf.memory.memory_reservation import opaque_memory_usage
+from rapidsmpf.memory.packed_data import PackedData
+from rapidsmpf.streaming.coll.allgather import AllGather
+from rapidsmpf.streaming.core.message import Message
 
 import pylibcudf as plc
 import rmm.mr
diff --git a/python/cudf_polars/pyproject.toml b/python/cudf_polars/pyproject.toml
index 3f395e4d92ee..b41121e7cbc6 100644
--- a/python/cudf_polars/pyproject.toml
+++ b/python/cudf_polars/pyproject.toml
@@ -27,6 +27,7 @@ dependencies = [
     "polars>=1.35,<1.42",
     "pylibcudf==26.8.*,>=0.0.0a0",
     "rapidsmpf==26.8.*,>=0.0.0a0",
+    "cudf_streaming==26.8.*,>=0.0.0a0",
     "typing_extensions>=4.0.0",
 ] # This list was generated by `rapids-dependency-file-generator`. To make changes, edit ../../dependencies.yaml and run `rapids-dependency-file-generator`.
 classifiers = [

From 2f447f1850e4e581b1a9279710b300799c705a19 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Tue, 2 Jun 2026 01:27:30 +0000
Subject: [PATCH 02/14] refactor(cudf_polars): update tests to use
 cudf_streaming imports

---
 python/cudf_polars/tests/streaming/test_allgather.py | 2 +-
 python/cudf_polars/tests/streaming/test_metadata.py  | 4 ++--
 python/cudf_polars/tests/streaming/test_shuffler.py  | 4 ++--
 python/cudf_polars/tests/streaming/test_spilling.py  | 2 +-
 4 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/python/cudf_polars/tests/streaming/test_allgather.py b/python/cudf_polars/tests/streaming/test_allgather.py
index 48d84afdb00f..c32e43c6d9b7 100644
--- a/python/cudf_polars/tests/streaming/test_allgather.py
+++ b/python/cudf_polars/tests/streaming/test_allgather.py
@@ -8,7 +8,7 @@
 import asyncio
 from concurrent.futures import ThreadPoolExecutor
 
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
+from cudf_streaming.streaming.table_chunk import TableChunk
 
 import polars as pl
 
diff --git a/python/cudf_polars/tests/streaming/test_metadata.py b/python/cudf_polars/tests/streaming/test_metadata.py
index 5c555e3593ba..40f2fd47f95a 100644
--- a/python/cudf_polars/tests/streaming/test_metadata.py
+++ b/python/cudf_polars/tests/streaming/test_metadata.py
@@ -6,14 +6,14 @@
 from __future__ import annotations
 
 import pytest
-from rapidsmpf.streaming.cudf.channel_metadata import (
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
     OrderKey,
     OrderScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
+from cudf_streaming.streaming.table_chunk import TableChunk
 
 import polars as pl
 
diff --git a/python/cudf_polars/tests/streaming/test_shuffler.py b/python/cudf_polars/tests/streaming/test_shuffler.py
index 7f4fd46d15e1..65db29541f81 100644
--- a/python/cudf_polars/tests/streaming/test_shuffler.py
+++ b/python/cudf_polars/tests/streaming/test_shuffler.py
@@ -6,12 +6,12 @@
 import asyncio
 
 import pytest
-from rapidsmpf.streaming.cudf.channel_metadata import (
+from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
     Partitioning,
 )
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
+from cudf_streaming.streaming.table_chunk import TableChunk
 
 import polars as pl
 
diff --git a/python/cudf_polars/tests/streaming/test_spilling.py b/python/cudf_polars/tests/streaming/test_spilling.py
index e368bc3ac1f3..35d74b2f59ee 100644
--- a/python/cudf_polars/tests/streaming/test_spilling.py
+++ b/python/cudf_polars/tests/streaming/test_spilling.py
@@ -13,7 +13,7 @@
 from rapidsmpf.memory.pinned_memory_resource import is_pinned_memory_resources_supported
 from rapidsmpf.streaming.core.message import Message
 from rapidsmpf.streaming.core.spillable_messages import SpillableMessages
-from rapidsmpf.streaming.cudf.table_chunk import TableChunk
+from cudf_streaming.streaming.table_chunk import TableChunk
 
 import polars as pl
 

From 2ef7257203a95a54e72aedb92b4727c02f408799 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Thu, 4 Jun 2026 21:04:33 +0000
Subject: [PATCH 03/14] fix(cudf_polars): add cudf_streaming to
 dependencies.yaml, fix import order

---
 dependencies.yaml                             | 23 ++++++++++++++++++-
 python/cudf_polars/pyproject.toml             |  2 +-
 .../tests/streaming/test_spilling.py          |  2 +-
 3 files changed, 24 insertions(+), 3 deletions(-)

diff --git a/dependencies.yaml b/dependencies.yaml
index 25ed1071f03d..69b0a0480f6b 100644
--- a/dependencies.yaml
+++ b/dependencies.yaml
@@ -335,6 +335,7 @@ files:
     includes:
       - run_cudf_polars
       - depends_on_rapidsmpf
+      - depends_on_cudf_streaming
       - depends_on_pylibcudf
       - depends_on_cuda_python
   # TODO(26.10): Remove this alias in favor of dask
@@ -1414,7 +1415,27 @@ dependencies:
     common:
       - output_types: conda
         packages:
-          - cudf-streaming==26.8.*,>=0.0.0a0
+          - &cudf_streaming_unsuffixed cudf-streaming==26.8.*,>=0.0.0a0
+      - output_types: requirements
+        packages:
+          # pip recognizes the index as a global option for the requirements.txt file
+          - --extra-index-url=https://pypi.anaconda.org/rapidsai-wheels-nightly/simple
+    specific:
+      - output_types: [requirements, pyproject]
+        matrices:
+          - matrix:
+              cuda: "12.*"
+              cuda_suffixed: "true"
+            packages:
+              - cudf-streaming-cu12==26.8.*,>=0.0.0a0
+          - matrix:
+              cuda: "13.*"
+              cuda_suffixed: "true"
+            packages:
+              - cudf-streaming-cu13==26.8.*,>=0.0.0a0
+          - matrix:
+            packages:
+              - *cudf_streaming_unsuffixed
   depends_on_ray:
     common:
       - output_types: conda
diff --git a/python/cudf_polars/pyproject.toml b/python/cudf_polars/pyproject.toml
index b41121e7cbc6..9b1ff8d7fd2f 100644
--- a/python/cudf_polars/pyproject.toml
+++ b/python/cudf_polars/pyproject.toml
@@ -22,12 +22,12 @@ license-files = ["LICENSE"]
 requires-python = ">=3.11"
 dependencies = [
     "cuda-python>=13.0.1,<14.0",
+    "cudf-streaming==26.8.*,>=0.0.0a0",
     "nvidia-ml-py>=12",
     "packaging",
     "polars>=1.35,<1.42",
     "pylibcudf==26.8.*,>=0.0.0a0",
     "rapidsmpf==26.8.*,>=0.0.0a0",
-    "cudf_streaming==26.8.*,>=0.0.0a0",
     "typing_extensions>=4.0.0",
 ] # This list was generated by `rapids-dependency-file-generator`. To make changes, edit ../../dependencies.yaml and run `rapids-dependency-file-generator`.
 classifiers = [
diff --git a/python/cudf_polars/tests/streaming/test_spilling.py b/python/cudf_polars/tests/streaming/test_spilling.py
index 35d74b2f59ee..9db08020aed3 100644
--- a/python/cudf_polars/tests/streaming/test_spilling.py
+++ b/python/cudf_polars/tests/streaming/test_spilling.py
@@ -9,11 +9,11 @@
 from typing import TYPE_CHECKING
 
 import pytest
+from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.memory.buffer import MemoryType
 from rapidsmpf.memory.pinned_memory_resource import is_pinned_memory_resources_supported
 from rapidsmpf.streaming.core.message import Message
 from rapidsmpf.streaming.core.spillable_messages import SpillableMessages
-from cudf_streaming.streaming.table_chunk import TableChunk
 
 import polars as pl
 

From 3021f597ea9db49718dec27aa2cf487089f810c2 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Thu, 4 Jun 2026 22:36:51 +0000
Subject: [PATCH 04/14] Download cudf-streaming wheels in cudf-polars CI tests

Add cudf-streaming and libcudf-streaming wheel downloads to the
cudf-polars wheel test scripts so they are installed from CI artifacts
rather than resolved from a remote index.

- Add wheel-build-cudf-streaming to needs for wheel-tests-cudf-polars
  and cudf-polars-polars-tests in pr.yaml
- Download libcudf_streaming and cudf_streaming wheels in
  test_wheel_cudf_polars.sh and test_cudf_polars_polars_tests.sh
- Add cudf-streaming as a runtime dependency in the cudf-polars
  conda recipe for conda-based test jobs (docs, narwhals)
---
 .github/workflows/pr.yaml             | 4 ++--
 ci/test_cudf_polars_polars_tests.sh   | 8 +++++++-
 ci/test_wheel_cudf_polars.sh          | 6 ++++++
 conda/recipes/cudf-polars/recipe.yaml | 1 +
 4 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index 031e26e4a048..ef239ac33acc 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -646,7 +646,7 @@ jobs:
       package-type: python
       pure-wheel: true
   wheel-tests-cudf-polars:
-    needs: [wheel-build-cudf-polars, changed-files]
+    needs: [wheel-build-cudf-polars, wheel-build-cudf-streaming, changed-files]
     permissions:
       actions: read
       contents: read
@@ -663,7 +663,7 @@ jobs:
       container-options: "--cap-add CAP_SYS_PTRACE --shm-size=8g --ulimit=nofile=1000000:1000000"
       script: "ci/test_wheel_cudf_polars.sh"
   cudf-polars-polars-tests:
-    needs: [wheel-build-cudf-polars, changed-files]
+    needs: [wheel-build-cudf-polars, wheel-build-cudf-streaming, changed-files]
     permissions:
       actions: read
       contents: read
diff --git a/ci/test_cudf_polars_polars_tests.sh b/ci/test_cudf_polars_polars_tests.sh
index 1fb9a4e395f6..331b85f2cc90 100755
--- a/ci/test_cudf_polars_polars_tests.sh
+++ b/ci/test_cudf_polars_polars_tests.sh
@@ -13,6 +13,10 @@ CUDF_POLARS_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cudf_polars_${RAPIDS_PY_CUDA_SUFF
 LIBCUDF_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 PYLIBCUDF_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" pylibcudf --stable --cuda "$RAPIDS_CUDA_VERSION")")
 
+# Download libcudf_streaming and cudf_streaming built in the previous step
+LIBCUDF_STREAMING_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_streaming_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
+CUDF_STREAMING_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" cudf_streaming --stable --cuda "$RAPIDS_CUDA_VERSION")")
+
 # generate constraints (possibly pinning to oldest support versions of dependencies)
 rapids-generate-pip-constraints py_test_cudf_polars "${PIP_CONSTRAINT}"
 
@@ -29,7 +33,9 @@ rapids-pip-retry install \
     --constraint "${PIP_CONSTRAINT}" \
     "$(echo "${CUDF_POLARS_WHEELHOUSE}"/cudf_polars_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)[test]" \
     "$(echo "${LIBCUDF_WHEELHOUSE}"/libcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
-    "$(echo "${PYLIBCUDF_WHEELHOUSE}"/pylibcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)"
+    "$(echo "${PYLIBCUDF_WHEELHOUSE}"/pylibcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
+    "$(echo "${LIBCUDF_STREAMING_WHEELHOUSE}"/libcudf_streaming_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
+    "$(echo "${CUDF_STREAMING_WHEELHOUSE}"/cudf_streaming_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)"
 
 TAG=$(python -c 'import polars; print(f"py-{polars.__version__}")')
 rapids-logger "Clone polars to ${TAG}"
diff --git a/ci/test_wheel_cudf_polars.sh b/ci/test_wheel_cudf_polars.sh
index 0b73e904dc47..61185c21568a 100755
--- a/ci/test_wheel_cudf_polars.sh
+++ b/ci/test_wheel_cudf_polars.sh
@@ -15,6 +15,10 @@ CUDF_POLARS_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cudf_polars_${RAPIDS_PY_CUDA_SUFF
 LIBCUDF_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 PYLIBCUDF_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" pylibcudf --stable --cuda "$RAPIDS_CUDA_VERSION")")
 
+# Download libcudf_streaming and cudf_streaming built in the previous step
+LIBCUDF_STREAMING_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_streaming_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
+CUDF_STREAMING_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" cudf_streaming --stable --cuda "$RAPIDS_CUDA_VERSION")")
+
 # generate constraints (possibly pinning to oldest support versions of dependencies)
 rapids-generate-pip-constraints py_test_cudf_polars "${PIP_CONSTRAINT}"
 
@@ -65,6 +69,8 @@ for version in "${VERSIONS[@]}"; do
         "$(echo "${CUDF_POLARS_WHEELHOUSE}"/cudf_polars_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)[test,dask,ray]" \
         "$(echo "${LIBCUDF_WHEELHOUSE}"/libcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
         "$(echo "${PYLIBCUDF_WHEELHOUSE}"/pylibcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
+        "$(echo "${LIBCUDF_STREAMING_WHEELHOUSE}"/libcudf_streaming_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
+        "$(echo "${CUDF_STREAMING_WHEELHOUSE}"/cudf_streaming_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
         -r "polars-compat-${version}-requirements.txt"
 
     rapids-logger "Running tests for polars ${version}.*"
diff --git a/conda/recipes/cudf-polars/recipe.yaml b/conda/recipes/cudf-polars/recipe.yaml
index 79ebec2963cb..12aa1e19f162 100644
--- a/conda/recipes/cudf-polars/recipe.yaml
+++ b/conda/recipes/cudf-polars/recipe.yaml
@@ -37,6 +37,7 @@ requirements:
     - python
     - pylibcudf =${{ version }}
     - rapidsmpf =${{ minor_version }}
+    - cudf-streaming =${{ version }}
     - polars>=1.35,<1.42
     - packaging
     - ${{ pin_compatible("cuda-version", upper_bound="x", lower_bound="x") }}

From 75f17afced56859595901778997c8ba8564de6a5 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Thu, 4 Jun 2026 23:37:19 +0000
Subject: [PATCH 05/14] Minor cleanup

---
 ci/test_cudf_polars_polars_tests.sh | 2 --
 1 file changed, 2 deletions(-)

diff --git a/ci/test_cudf_polars_polars_tests.sh b/ci/test_cudf_polars_polars_tests.sh
index 331b85f2cc90..50d1ee861f16 100755
--- a/ci/test_cudf_polars_polars_tests.sh
+++ b/ci/test_cudf_polars_polars_tests.sh
@@ -12,8 +12,6 @@ RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen "${RAPIDS_CUDA_VERSION}")"
 CUDF_POLARS_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cudf_polars_${RAPIDS_PY_CUDA_SUFFIX}" RAPIDS_PY_WHEEL_PURE="1" rapids-download-wheels-from-github python)
 LIBCUDF_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 PYLIBCUDF_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" pylibcudf --stable --cuda "$RAPIDS_CUDA_VERSION")")
-
-# Download libcudf_streaming and cudf_streaming built in the previous step
 LIBCUDF_STREAMING_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_streaming_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
 CUDF_STREAMING_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" cudf_streaming --stable --cuda "$RAPIDS_CUDA_VERSION")")
 

From 7dea206a7307333503ef2be57c4d946ea7bc627d Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Thu, 4 Jun 2026 23:55:12 +0000
Subject: [PATCH 06/14] Remove workaround for building rapidsmpf in
 devcontainer builds

---
 .github/workflows/pr.yaml | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index ef239ac33acc..1a345e4a498f 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -732,14 +732,9 @@ jobs:
         SCCACHE_DIST_MAX_RETRIES=inf
         SCCACHE_SERVER_LOG=sccache=debug
         SCCACHE_DIST_FALLBACK_TO_LOCAL_COMPILE=false
-      # Call clone-rapidsmpf before creating the Python env to exclude cudf_polars' dependency on rapidsmpf
-      # librapidsmpf-cu13 wheels brings in a hardcoded libnuma-dev cmake target: https://github.com/NVIDIA/cuCascade/issues/118
-      # -DBUILD_TESTS=OFF to match rapidsmpf https://github.com/rapidsai/rapidsmpf/blob/main/.github/workflows/pr.yaml#L351 (leads to compilation errors)
-      # -DCUDF_BUILD_TESTUTIL=OFF to avoid IMPORTED_GLOBAL promotion errors when cuCascade's find_package(cudf) loads cudf-config.cmake from a CPM subdirectory
       build_command: |
         sccache --zero-stats;
-        clone-rapidsmpf -j$(nproc) -v -q --branch "$(cat ~/cudf/RAPIDS_BRANCH)" --clone-upstream --depth 1 --single-branch --shallow-submodules;
-        build-all -j0 -DBUILD_BENCHMARKS=OFF -DBUILD_NUMA_SUPPORT=OFF -DBUILD_TESTS=OFF -DCUDF_BUILD_TESTUTIL=OFF --verbose 2>&1 | tee telemetry-artifacts/build.log;
+        build-all -j0 -DBUILD_BENCHMARKS=ON --verbose 2>&1 | tee telemetry-artifacts/build.log;
         sccache --show-adv-stats | tee telemetry-artifacts/sccache-stats.txt;
   unit-tests-cudf-pandas:
     needs: [wheel-build-cudf, changed-files]

From 3efe63b2f37f3f41d2bdb205922c3c4ca9a28a7c Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Fri, 5 Jun 2026 00:06:09 +0000
Subject: [PATCH 07/14] Update isort sections

---
 python/cudf_polars/pyproject.toml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/cudf_polars/pyproject.toml b/python/cudf_polars/pyproject.toml
index 9b1ff8d7fd2f..557520d9a2fe 100644
--- a/python/cudf_polars/pyproject.toml
+++ b/python/cudf_polars/pyproject.toml
@@ -233,7 +233,7 @@ required-imports = ["from __future__ import annotations"]
 
 [tool.ruff.lint.isort.sections]
 polars = ["polars"]
-rapids = ["rmm", "pylibcudf"]
+rapids = ["rmm", "pylibcudf", "rapidsmpf", "cudf_streaming"]
 
 [tool.ruff.format]
 docstring-code-format = true

From 5b9fac413ee00084b710e52b093d264ddb9fdb7f Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Fri, 5 Jun 2026 00:12:11 +0000
Subject: [PATCH 08/14] Apply isort fix

---
 python/cudf_polars/cudf_polars/engine/core.py         |  5 +++--
 python/cudf_polars/cudf_polars/engine/dask.py         |  8 ++++----
 python/cudf_polars/cudf_polars/engine/ray.py          | 11 ++++++-----
 python/cudf_polars/cudf_polars/engine/spmd.py         |  9 ++++-----
 .../streaming/actor_graph/collectives/allgather.py    |  6 ++----
 .../streaming/actor_graph/collectives/shuffle.py      |  6 ++----
 .../streaming/actor_graph/collectives/sort.py         |  7 +++----
 .../cudf_polars/streaming/actor_graph/core.py         |  4 ++--
 .../cudf_polars/streaming/actor_graph/groupby.py      |  3 +--
 .../cudf_polars/streaming/actor_graph/io.py           |  3 +--
 .../cudf_polars/streaming/actor_graph/join.py         |  5 ++---
 .../cudf_polars/streaming/actor_graph/over.py         |  8 +++-----
 .../cudf_polars/streaming/actor_graph/utils.py        |  6 ++----
 python/cudf_polars/cudf_polars/utils/config.py        |  4 ++--
 python/cudf_polars/tests/streaming/test_allgather.py  |  3 +--
 python/cudf_polars/tests/streaming/test_dask.py       |  3 ++-
 python/cudf_polars/tests/streaming/test_metadata.py   |  8 ++++----
 python/cudf_polars/tests/streaming/test_ray.py        |  3 ++-
 python/cudf_polars/tests/streaming/test_shuffler.py   |  5 +++--
 python/cudf_polars/tests/streaming/test_spilling.py   |  8 ++++----
 python/cudf_polars/tests/streaming/test_spmd.py       |  4 ++--
 python/cudf_polars/tests/streaming/test_statistics.py |  1 +
 22 files changed, 56 insertions(+), 64 deletions(-)

diff --git a/python/cudf_polars/cudf_polars/engine/core.py b/python/cudf_polars/cudf_polars/engine/core.py
index f7881b5ea95a..bfe97050ccf3 100644
--- a/python/cudf_polars/cudf_polars/engine/core.py
+++ b/python/cudf_polars/cudf_polars/engine/core.py
@@ -14,6 +14,9 @@
 from typing import TYPE_CHECKING, Any, ClassVar, Self, TypeVar
 
 import cuda.core
+
+import polars as pl
+
 from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.coll import AllGather
 from rapidsmpf.config import Options, get_environment_variables
@@ -21,8 +24,6 @@
 from rapidsmpf.statistics import Statistics
 from rapidsmpf.streaming.core.actor import run_actor_network
 
-import polars as pl
-
 from cudf_polars.containers import DataFrame
 from cudf_polars.dsl.ir import IRExecutionContext
 from cudf_polars.streaming.actor_graph.collectives import ReserveOpIDs
diff --git a/python/cudf_polars/cudf_polars/engine/dask.py b/python/cudf_polars/cudf_polars/engine/dask.py
index f37a5db27998..46372d0dad04 100644
--- a/python/cudf_polars/cudf_polars/engine/dask.py
+++ b/python/cudf_polars/cudf_polars/engine/dask.py
@@ -16,6 +16,10 @@
 import distributed.system
 import pynvml
 import ucxx._lib.libucxx as ucx_api
+
+import polars as pl
+
+import rmm.mr
 from rapidsmpf import bootstrap
 from rapidsmpf.communicator.ucxx import barrier, get_root_ucxx_address, new_communicator
 from rapidsmpf.config import Options
@@ -24,10 +28,6 @@
 from rapidsmpf.statistics import Statistics
 from rapidsmpf.streaming.core.context import Context
 
-import polars as pl
-
-import rmm.mr
-
 from cudf_polars.engine.core import (
     ClusterInfo,
     StreamingEngine,
diff --git a/python/cudf_polars/cudf_polars/engine/ray.py b/python/cudf_polars/cudf_polars/engine/ray.py
index 686767230e17..b8824ee55414 100644
--- a/python/cudf_polars/cudf_polars/engine/ray.py
+++ b/python/cudf_polars/cudf_polars/engine/ray.py
@@ -12,6 +12,10 @@
 
 import ray
 import ucxx._lib.libucxx as ucx_api
+
+import polars as pl
+
+import rmm.mr
 from rapidsmpf import bootstrap
 from rapidsmpf.communicator.ucxx import barrier, get_root_ucxx_address, new_communicator
 from rapidsmpf.config import Options
@@ -20,10 +24,6 @@
 from rapidsmpf.statistics import Statistics
 from rapidsmpf.streaming.core.context import Context
 
-import polars as pl
-
-import rmm.mr
-
 from cudf_polars.engine.core import (
     ClusterInfo,
     StreamingEngine,
@@ -41,9 +41,10 @@
     import uuid
     from collections.abc import Callable
 
+    from ray.actor import ActorHandle
+
     from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from rapidsmpf.communicator.communicator import Communicator
-    from ray.actor import ActorHandle
 
     from cudf_polars.dsl.ir import IR
     from cudf_polars.engine.core import T
diff --git a/python/cudf_polars/cudf_polars/engine/spmd.py b/python/cudf_polars/cudf_polars/engine/spmd.py
index c6f057d48b12..e7905c845c56 100644
--- a/python/cudf_polars/cudf_polars/engine/spmd.py
+++ b/python/cudf_polars/cudf_polars/engine/spmd.py
@@ -10,7 +10,9 @@
 from concurrent.futures import ThreadPoolExecutor
 from typing import TYPE_CHECKING, Any, cast
 
+import pylibcudf as plc
 from cudf_streaming.integrations.partition import unpack_and_concat
+from pylibcudf.contiguous_split import pack
 from rapidsmpf import bootstrap
 from rapidsmpf.coll import AllGather
 from rapidsmpf.communicator.single import (
@@ -23,9 +25,6 @@
 from rapidsmpf.statistics import Statistics
 from rapidsmpf.streaming.core.context import Context
 
-import pylibcudf as plc
-from pylibcudf.contiguous_split import pack
-
 from cudf_polars.containers import DataFrame, DataType
 from cudf_polars.engine.core import (
     ClusterInfo,
@@ -51,12 +50,12 @@
     import uuid
     from collections.abc import Callable
 
+    import polars as pl
+
     from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.config import Options
 
-    import polars as pl
-
     from cudf_polars.dsl.ir import IR
     from cudf_polars.engine.core import T
     from cudf_polars.engine.options import StreamingOptions
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
index 5e899958d95f..1772e9bdb41e 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
@@ -7,17 +7,15 @@
 from typing import TYPE_CHECKING
 
 from cudf_streaming.integrations.partition import unpack_and_concat
+from pylibcudf.contiguous_split import pack
 from rapidsmpf.memory.packed_data import PackedData
 from rapidsmpf.streaming.coll.allgather import AllGather
 
-from pylibcudf.contiguous_split import pack
-
 if TYPE_CHECKING:
+    import pylibcudf as plc
     from cudf_streaming.streaming.table_chunk import TableChunk
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.streaming.core.context import Context
-
-    import pylibcudf as plc
     from rmm.pylibrmm.stream import Stream
 
     from cudf_polars.dsl.ir import IRExecutionContext
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py
index 99c2f38d5069..dd78104d2e96 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/shuffle.py
@@ -6,6 +6,8 @@
 
 from typing import TYPE_CHECKING, Any
 
+import pylibcudf as plc
+import pylibcudf.partitioning
 from cudf_streaming.integrations.partition import (
     partition_and_pack as py_partition_and_pack,
     split_and_pack as py_split_and_pack,
@@ -25,9 +27,6 @@
 from rapidsmpf.streaming.core.context import Context
 from rapidsmpf.streaming.core.message import Message
 
-import pylibcudf as plc
-import pylibcudf.partitioning
-
 from cudf_polars.dsl.expr import Col
 from cudf_polars.streaming.actor_graph.dispatch import (
     generate_ir_sub_network,
@@ -48,7 +47,6 @@
     from rapidsmpf.communicator.communicator import Communicator
     from rapidsmpf.memory.packed_data import PackedData
     from rapidsmpf.streaming.core.channel import Channel
-
     from rmm.pylibrmm.stream import Stream
 
     from cudf_polars.dsl.ir import IR, IRExecutionContext
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py
index cec4901d5641..f62f2fa05f76 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/sort.py
@@ -6,6 +6,9 @@
 
 from typing import TYPE_CHECKING
 
+import polars as pl
+
+import pylibcudf as plc
 from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     OrderKey,
@@ -17,10 +20,6 @@
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.message import Message
 
-import polars as pl
-
-import pylibcudf as plc
-
 from cudf_polars.containers import DataFrame, DataType
 from cudf_polars.dsl.expr import Col, NamedExpr
 from cudf_polars.dsl.ir import Empty, Sort
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py
index 0160681262a1..1bd440a239f5 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/core.py
@@ -30,6 +30,8 @@
 if TYPE_CHECKING:
     from collections.abc import MutableMapping
 
+    import polars as pl
+
     from cudf_streaming.streaming.channel_metadata import ChannelMetadata
     from cudf_streaming.streaming.table_chunk import TableChunk
     from rapidsmpf.communicator.communicator import Communicator
@@ -37,8 +39,6 @@
     from rapidsmpf.streaming.core.context import Context
     from rapidsmpf.streaming.core.leaf_actor import DeferredMessages
 
-    import polars as pl
-
     from cudf_polars.dsl.ir import IR, IRExecutionContext
     from cudf_polars.streaming.actor_graph.dispatch import (
         GenState,
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py
index 932c3662b99b..df2c8d04f4c7 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/groupby.py
@@ -7,6 +7,7 @@
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any
 
+import pylibcudf as plc
 from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
@@ -18,8 +19,6 @@
 from rapidsmpf.streaming.core.context import Context
 from rapidsmpf.streaming.core.message import Message
 
-import pylibcudf as plc
-
 from cudf_polars.containers import DataType
 from cudf_polars.dsl.expr import Col, NamedExpr
 from cudf_polars.dsl.ir import IR, Distinct, GroupBy, Select
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
index 8abf0a25c9fa..f30b55dbdd3d 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/io.py
@@ -8,6 +8,7 @@
 import math
 from typing import TYPE_CHECKING, Any
 
+import pylibcudf as plc
 from cudf_streaming.streaming.channel_metadata import ChannelMetadata
 from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.memory.memory_reservation import opaque_memory_usage
@@ -16,8 +17,6 @@
 )
 from rapidsmpf.streaming.core.message import Message
 
-import pylibcudf as plc
-
 from cudf_polars.dsl.ir import (
     IR,
     DataFrameScan,
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py
index 8cac2a693bac..fa99192aaf12 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/join.py
@@ -7,6 +7,7 @@
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any, Literal
 
+import pylibcudf as plc
 from cudf_streaming.streaming.bloom_filter import BloomFilter
 from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
@@ -17,6 +18,7 @@
     TableChunk,
     make_table_chunks_available_or_wait,
 )
+from pylibcudf.hashing import LIBCUDF_DEFAULT_HASH_SEED
 from rapidsmpf.memory.memory_reservation import opaque_memory_usage
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.memory_reserve_or_wait import (
@@ -24,9 +26,6 @@
 )
 from rapidsmpf.streaming.core.message import Message
 
-import pylibcudf as plc
-from pylibcudf.hashing import LIBCUDF_DEFAULT_HASH_SEED
-
 from cudf_polars.containers import DataFrame
 from cudf_polars.dsl.ir import IR, Join
 from cudf_polars.dsl.utils.naming import names_to_indices
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
index 623e1c9c5cc1..4b0f20db5f7b 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/over.py
@@ -40,6 +40,9 @@
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Any, ClassVar, cast
 
+import polars as pl
+
+import pylibcudf as plc
 from cudf_streaming.streaming.channel_metadata import ChannelMetadata
 from cudf_streaming.streaming.table_chunk import (
     TableChunk,
@@ -50,10 +53,6 @@
 from rapidsmpf.streaming.core.actor import define_actor
 from rapidsmpf.streaming.core.message import Message
 
-import polars as pl
-
-import pylibcudf as plc
-
 from cudf_polars.containers import Column, DataFrame, DataType
 from cudf_polars.dsl.expr import GroupedWindow
 from cudf_polars.dsl.expressions.base import ExecutionContext
@@ -92,7 +91,6 @@
     from rapidsmpf.memory.buffer_resource import BufferResource
     from rapidsmpf.streaming.core.channel import Channel
     from rapidsmpf.streaming.core.context import Context
-
     from rmm.pylibrmm.stream import Stream
 
     from cudf_polars.dsl.expr import Col
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py
index 1679391574ee..68c5e8a94c0d 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/utils.py
@@ -16,6 +16,8 @@
 from functools import reduce
 from typing import TYPE_CHECKING, Any, Literal, TypeAlias, cast
 
+import pylibcudf as plc
+import rmm.mr
 from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
@@ -32,9 +34,6 @@
 from rapidsmpf.streaming.coll.allgather import AllGather
 from rapidsmpf.streaming.core.message import Message
 
-import pylibcudf as plc
-import rmm.mr
-
 import cudf_polars.dsl.tracing
 from cudf_polars.containers import DataFrame
 from cudf_polars.dsl.expr import Col, NamedExpr
@@ -61,7 +60,6 @@
     from rapidsmpf.streaming.core.channel import Channel
     from rapidsmpf.streaming.core.context import Context
     from rapidsmpf.streaming.core.spillable_messages import SpillableMessages
-
     from rmm.pylibrmm.stream import Stream
 
     from cudf_polars.dsl.ir import IR, IRExecutionContext
diff --git a/python/cudf_polars/cudf_polars/utils/config.py b/python/cudf_polars/cudf_polars/utils/config.py
index 4582c0040019..5786a5351cc4 100644
--- a/python/cudf_polars/cudf_polars/utils/config.py
+++ b/python/cudf_polars/cudf_polars/utils/config.py
@@ -36,13 +36,13 @@
     from concurrent.futures import ThreadPoolExecutor
 
     import distributed
-    from rapidsmpf.communicator.communicator import Communicator
-    from rapidsmpf.streaming.core.context import Context
     from ray.actor import ActorHandle
 
     import polars.lazyframe.engine_config
 
     import rmm.mr
+    from rapidsmpf.communicator.communicator import Communicator
+    from rapidsmpf.streaming.core.context import Context
 
     from cudf_polars.engine.ray import RankActor
 
diff --git a/python/cudf_polars/tests/streaming/test_allgather.py b/python/cudf_polars/tests/streaming/test_allgather.py
index c32e43c6d9b7..54cd49b9afff 100644
--- a/python/cudf_polars/tests/streaming/test_allgather.py
+++ b/python/cudf_polars/tests/streaming/test_allgather.py
@@ -8,11 +8,10 @@
 import asyncio
 from concurrent.futures import ThreadPoolExecutor
 
-from cudf_streaming.streaming.table_chunk import TableChunk
-
 import polars as pl
 
 import pylibcudf as plc
+from cudf_streaming.streaming.table_chunk import TableChunk
 
 from cudf_polars.dsl.ir import IRExecutionContext
 from cudf_polars.streaming.actor_graph.collectives.allgather import AllGatherManager
diff --git a/python/cudf_polars/tests/streaming/test_dask.py b/python/cudf_polars/tests/streaming/test_dask.py
index 3d81ff5cf1c8..c5cabd323aa8 100644
--- a/python/cudf_polars/tests/streaming/test_dask.py
+++ b/python/cudf_polars/tests/streaming/test_dask.py
@@ -9,10 +9,11 @@
 from typing import TYPE_CHECKING
 
 import pytest
-from rapidsmpf.bootstrap import is_running_with_rrun
 
 import polars as pl
 
+from rapidsmpf.bootstrap import is_running_with_rrun
+
 from cudf_polars.engine.options import StreamingOptions
 from cudf_polars.testing.asserts import assert_gpu_result_equal
 from cudf_polars.utils.config import DaskContext
diff --git a/python/cudf_polars/tests/streaming/test_metadata.py b/python/cudf_polars/tests/streaming/test_metadata.py
index 40f2fd47f95a..d634c88561a3 100644
--- a/python/cudf_polars/tests/streaming/test_metadata.py
+++ b/python/cudf_polars/tests/streaming/test_metadata.py
@@ -6,6 +6,10 @@
 from __future__ import annotations
 
 import pytest
+
+import polars as pl
+
+import pylibcudf as plc
 from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
@@ -15,10 +19,6 @@
 )
 from cudf_streaming.streaming.table_chunk import TableChunk
 
-import polars as pl
-
-import pylibcudf as plc
-
 from cudf_polars import Translator
 from cudf_polars.containers import DataFrame, DataType
 from cudf_polars.dsl import expr
diff --git a/python/cudf_polars/tests/streaming/test_ray.py b/python/cudf_polars/tests/streaming/test_ray.py
index a4e543d5c628..3b226eda8147 100644
--- a/python/cudf_polars/tests/streaming/test_ray.py
+++ b/python/cudf_polars/tests/streaming/test_ray.py
@@ -9,10 +9,11 @@
 from unittest.mock import patch
 
 import pytest
-from rapidsmpf.bootstrap import is_running_with_rrun
 
 import polars as pl
 
+from rapidsmpf.bootstrap import is_running_with_rrun
+
 from cudf_polars.engine.hardware_binding import HardwareBindingPolicy
 from cudf_polars.utils.config import RayContext
 
diff --git a/python/cudf_polars/tests/streaming/test_shuffler.py b/python/cudf_polars/tests/streaming/test_shuffler.py
index 65db29541f81..51ffe28e7f46 100644
--- a/python/cudf_polars/tests/streaming/test_shuffler.py
+++ b/python/cudf_polars/tests/streaming/test_shuffler.py
@@ -6,6 +6,9 @@
 import asyncio
 
 import pytest
+
+import polars as pl
+
 from cudf_streaming.streaming.channel_metadata import (
     ChannelMetadata,
     HashScheme,
@@ -13,8 +16,6 @@
 )
 from cudf_streaming.streaming.table_chunk import TableChunk
 
-import polars as pl
-
 from cudf_polars.containers import DataFrame, DataType
 from cudf_polars.engine.options import StreamingOptions
 from cudf_polars.engine.spmd import allgather_polars_dataframe
diff --git a/python/cudf_polars/tests/streaming/test_spilling.py b/python/cudf_polars/tests/streaming/test_spilling.py
index 9db08020aed3..c1d7e5b1cdfc 100644
--- a/python/cudf_polars/tests/streaming/test_spilling.py
+++ b/python/cudf_polars/tests/streaming/test_spilling.py
@@ -9,16 +9,16 @@
 from typing import TYPE_CHECKING
 
 import pytest
+
+import polars as pl
+
+import pylibcudf as plc
 from cudf_streaming.streaming.table_chunk import TableChunk
 from rapidsmpf.memory.buffer import MemoryType
 from rapidsmpf.memory.pinned_memory_resource import is_pinned_memory_resources_supported
 from rapidsmpf.streaming.core.message import Message
 from rapidsmpf.streaming.core.spillable_messages import SpillableMessages
 
-import polars as pl
-
-import pylibcudf as plc
-
 from cudf_polars.engine.options import StreamingOptions
 from cudf_polars.streaming.actor_graph.utils import (
     make_spill_function,
diff --git a/python/cudf_polars/tests/streaming/test_spmd.py b/python/cudf_polars/tests/streaming/test_spmd.py
index 17c2f86d961a..ed4a5b069ba9 100644
--- a/python/cudf_polars/tests/streaming/test_spmd.py
+++ b/python/cudf_polars/tests/streaming/test_spmd.py
@@ -9,12 +9,12 @@
 from unittest.mock import patch
 
 import pytest
-from rapidsmpf.bootstrap import is_running_with_rrun
-from rapidsmpf.rmm_resource_adaptor import RmmResourceAdaptor
 
 import polars as pl
 
 import rmm.mr
+from rapidsmpf.bootstrap import is_running_with_rrun
+from rapidsmpf.rmm_resource_adaptor import RmmResourceAdaptor
 
 from cudf_polars.engine.core import _find_memory_error
 from cudf_polars.engine.hardware_binding import HardwareBindingPolicy
diff --git a/python/cudf_polars/tests/streaming/test_statistics.py b/python/cudf_polars/tests/streaming/test_statistics.py
index 61d5659dc702..f51605d6bfd7 100644
--- a/python/cudf_polars/tests/streaming/test_statistics.py
+++ b/python/cudf_polars/tests/streaming/test_statistics.py
@@ -7,6 +7,7 @@
 from typing import TYPE_CHECKING
 
 import pytest
+
 from rapidsmpf.statistics import Statistics
 
 from cudf_polars.engine.options import StreamingOptions

From f85db908dff7b935f285636bffc0a2325d4f6a99 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Mon, 8 Jun 2026 15:16:49 +0000
Subject: [PATCH 09/14] Revert "Remove workaround for building rapidsmpf in
 devcontainer builds"

This reverts commit 7dea206a7307333503ef2be57c4d946ea7bc627d.
---
 .github/workflows/pr.yaml | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index 1a345e4a498f..ef239ac33acc 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -732,9 +732,14 @@ jobs:
         SCCACHE_DIST_MAX_RETRIES=inf
         SCCACHE_SERVER_LOG=sccache=debug
         SCCACHE_DIST_FALLBACK_TO_LOCAL_COMPILE=false
+      # Call clone-rapidsmpf before creating the Python env to exclude cudf_polars' dependency on rapidsmpf
+      # librapidsmpf-cu13 wheels brings in a hardcoded libnuma-dev cmake target: https://github.com/NVIDIA/cuCascade/issues/118
+      # -DBUILD_TESTS=OFF to match rapidsmpf https://github.com/rapidsai/rapidsmpf/blob/main/.github/workflows/pr.yaml#L351 (leads to compilation errors)
+      # -DCUDF_BUILD_TESTUTIL=OFF to avoid IMPORTED_GLOBAL promotion errors when cuCascade's find_package(cudf) loads cudf-config.cmake from a CPM subdirectory
       build_command: |
         sccache --zero-stats;
-        build-all -j0 -DBUILD_BENCHMARKS=ON --verbose 2>&1 | tee telemetry-artifacts/build.log;
+        clone-rapidsmpf -j$(nproc) -v -q --branch "$(cat ~/cudf/RAPIDS_BRANCH)" --clone-upstream --depth 1 --single-branch --shallow-submodules;
+        build-all -j0 -DBUILD_BENCHMARKS=OFF -DBUILD_NUMA_SUPPORT=OFF -DBUILD_TESTS=OFF -DCUDF_BUILD_TESTUTIL=OFF --verbose 2>&1 | tee telemetry-artifacts/build.log;
         sccache --show-adv-stats | tee telemetry-artifacts/sccache-stats.txt;
   unit-tests-cudf-pandas:
     needs: [wheel-build-cudf, changed-files]

From 3952f856bb244297ad118b589f4fb2b37f2606b6 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Mon, 8 Jun 2026 15:28:03 +0000
Subject: [PATCH 10/14] Reapply "Remove workaround for building rapidsmpf in
 devcontainer builds"

This reverts commit f85db908dff7b935f285636bffc0a2325d4f6a99.
---
 .github/workflows/pr.yaml                         | 7 +------
 conda/environments/all_cuda-129_arch-aarch64.yaml | 2 ++
 conda/environments/all_cuda-129_arch-x86_64.yaml  | 2 ++
 conda/environments/all_cuda-132_arch-aarch64.yaml | 2 ++
 conda/environments/all_cuda-132_arch-x86_64.yaml  | 2 ++
 dependencies.yaml                                 | 2 ++
 6 files changed, 11 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index ef239ac33acc..1a345e4a498f 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -732,14 +732,9 @@ jobs:
         SCCACHE_DIST_MAX_RETRIES=inf
         SCCACHE_SERVER_LOG=sccache=debug
         SCCACHE_DIST_FALLBACK_TO_LOCAL_COMPILE=false
-      # Call clone-rapidsmpf before creating the Python env to exclude cudf_polars' dependency on rapidsmpf
-      # librapidsmpf-cu13 wheels brings in a hardcoded libnuma-dev cmake target: https://github.com/NVIDIA/cuCascade/issues/118
-      # -DBUILD_TESTS=OFF to match rapidsmpf https://github.com/rapidsai/rapidsmpf/blob/main/.github/workflows/pr.yaml#L351 (leads to compilation errors)
-      # -DCUDF_BUILD_TESTUTIL=OFF to avoid IMPORTED_GLOBAL promotion errors when cuCascade's find_package(cudf) loads cudf-config.cmake from a CPM subdirectory
       build_command: |
         sccache --zero-stats;
-        clone-rapidsmpf -j$(nproc) -v -q --branch "$(cat ~/cudf/RAPIDS_BRANCH)" --clone-upstream --depth 1 --single-branch --shallow-submodules;
-        build-all -j0 -DBUILD_BENCHMARKS=OFF -DBUILD_NUMA_SUPPORT=OFF -DBUILD_TESTS=OFF -DCUDF_BUILD_TESTUTIL=OFF --verbose 2>&1 | tee telemetry-artifacts/build.log;
+        build-all -j0 -DBUILD_BENCHMARKS=ON --verbose 2>&1 | tee telemetry-artifacts/build.log;
         sccache --show-adv-stats | tee telemetry-artifacts/sccache-stats.txt;
   unit-tests-cudf-pandas:
     needs: [wheel-build-cudf, changed-files]
diff --git a/conda/environments/all_cuda-129_arch-aarch64.yaml b/conda/environments/all_cuda-129_arch-aarch64.yaml
index 838a2c557b47..7e0ad88ddb96 100644
--- a/conda/environments/all_cuda-129_arch-aarch64.yaml
+++ b/conda/environments/all_cuda-129_arch-aarch64.yaml
@@ -44,6 +44,7 @@ dependencies:
 - libkvikio==26.8.*,>=0.0.0a0
 - libnvcomp-dev==5.2.0.10
 - libnvjitlink-dev
+- librapidsmpf==26.8.*,>=0.0.0a0
 - librdkafka>=2.11.1,<2.12.0
 - librmm==26.8.*,>=0.0.0a0
 - make
@@ -87,6 +88,7 @@ dependencies:
 - rapids-build-backend>=0.4.0,<0.5.0
 - rapids-dask-dependency==26.8.*,>=0.0.0a0
 - rapids-logger==0.2.*,>=0.0.0a0
+- rapidsmpf==26.8.*,>=0.0.0a0
 - ray-default>=2.55.1
 - rich
 - rmm==26.8.*,>=0.0.0a0
diff --git a/conda/environments/all_cuda-129_arch-x86_64.yaml b/conda/environments/all_cuda-129_arch-x86_64.yaml
index 0a435504e32f..272914ad7635 100644
--- a/conda/environments/all_cuda-129_arch-x86_64.yaml
+++ b/conda/environments/all_cuda-129_arch-x86_64.yaml
@@ -44,6 +44,7 @@ dependencies:
 - libkvikio==26.8.*,>=0.0.0a0
 - libnvcomp-dev==5.2.0.10
 - libnvjitlink-dev
+- librapidsmpf==26.8.*,>=0.0.0a0
 - librdkafka>=2.11.1,<2.12.0
 - librmm==26.8.*,>=0.0.0a0
 - make
@@ -87,6 +88,7 @@ dependencies:
 - rapids-build-backend>=0.4.0,<0.5.0
 - rapids-dask-dependency==26.8.*,>=0.0.0a0
 - rapids-logger==0.2.*,>=0.0.0a0
+- rapidsmpf==26.8.*,>=0.0.0a0
 - ray-default>=2.55.1
 - rich
 - rmm==26.8.*,>=0.0.0a0
diff --git a/conda/environments/all_cuda-132_arch-aarch64.yaml b/conda/environments/all_cuda-132_arch-aarch64.yaml
index 2093888e01fa..9ecba408142c 100644
--- a/conda/environments/all_cuda-132_arch-aarch64.yaml
+++ b/conda/environments/all_cuda-132_arch-aarch64.yaml
@@ -44,6 +44,7 @@ dependencies:
 - libkvikio==26.8.*,>=0.0.0a0
 - libnvcomp-dev==5.2.0.10
 - libnvjitlink-dev
+- librapidsmpf==26.8.*,>=0.0.0a0
 - librdkafka>=2.11.1,<2.12.0
 - librmm==26.8.*,>=0.0.0a0
 - make
@@ -87,6 +88,7 @@ dependencies:
 - rapids-build-backend>=0.4.0,<0.5.0
 - rapids-dask-dependency==26.8.*,>=0.0.0a0
 - rapids-logger==0.2.*,>=0.0.0a0
+- rapidsmpf==26.8.*,>=0.0.0a0
 - ray-default>=2.55.1
 - rich
 - rmm==26.8.*,>=0.0.0a0
diff --git a/conda/environments/all_cuda-132_arch-x86_64.yaml b/conda/environments/all_cuda-132_arch-x86_64.yaml
index 360550447330..ced21db00730 100644
--- a/conda/environments/all_cuda-132_arch-x86_64.yaml
+++ b/conda/environments/all_cuda-132_arch-x86_64.yaml
@@ -44,6 +44,7 @@ dependencies:
 - libkvikio==26.8.*,>=0.0.0a0
 - libnvcomp-dev==5.2.0.10
 - libnvjitlink-dev
+- librapidsmpf==26.8.*,>=0.0.0a0
 - librdkafka>=2.11.1,<2.12.0
 - librmm==26.8.*,>=0.0.0a0
 - make
@@ -87,6 +88,7 @@ dependencies:
 - rapids-build-backend>=0.4.0,<0.5.0
 - rapids-dask-dependency==26.8.*,>=0.0.0a0
 - rapids-logger==0.2.*,>=0.0.0a0
+- rapidsmpf==26.8.*,>=0.0.0a0
 - ray-default>=2.55.1
 - rich
 - rmm==26.8.*,>=0.0.0a0
diff --git a/dependencies.yaml b/dependencies.yaml
index 69b0a0480f6b..0409d0c4cfe2 100644
--- a/dependencies.yaml
+++ b/dependencies.yaml
@@ -23,8 +23,10 @@ files:
       - depends_on_libkvikio
       - depends_on_librmm
       - depends_on_libnvcomp
+      - depends_on_librapidsmpf
       - depends_on_numba_cuda
       - depends_on_rapids_logger
+      - depends_on_rapidsmpf
       - depends_on_ray
       - depends_on_rmm
       - develop

From 9fb2ab80cddaf11f315f5a2f871a9a184bd97584 Mon Sep 17 00:00:00 2001
From: Bradley Dice <bdice@bradleydice.com>
Date: Mon, 8 Jun 2026 11:07:41 -0500
Subject: [PATCH 11/14] Remove rapidsmpf sibling mount from devcontainers

---
 .devcontainer/README.md                        | 8 --------
 .devcontainer/cuda12.9-conda/devcontainer.json | 5 ++---
 .devcontainer/cuda12.9-pip/devcontainer.json   | 5 ++---
 .devcontainer/cuda13.2-conda/devcontainer.json | 5 ++---
 .devcontainer/cuda13.2-pip/devcontainer.json   | 5 ++---
 5 files changed, 8 insertions(+), 20 deletions(-)

diff --git a/.devcontainer/README.md b/.devcontainer/README.md
index cc5fac22fde2..91ee7ef85f78 100644
--- a/.devcontainer/README.md
+++ b/.devcontainer/README.md
@@ -20,7 +20,6 @@ This container is a turnkey development environment for building and testing the
 By default, the following directories are bind-mounted into the devcontainer:
 
 * `${repo}:/home/coder/cudf`
-* `${repo}/../rapidsmpf:/home/coder/rapidsmpf`
 * `${repo}/../.aws:/home/coder/.aws`
 * `${repo}/../.local:/home/coder/.local`
 * `${repo}/../.cache:/home/coder/.cache`
@@ -29,13 +28,6 @@ By default, the following directories are bind-mounted into the devcontainer:
 
 This ensures caches, configurations, dependencies, and your commits are persisted on the host across container runs.
 
-The [rapidsmpf](https://github.com/rapidsai/rapidsmpf) repository is a required dependency of `cudf_polars` (that also requires `libcudf`) and must be cloned as a sibling directory to the cudf repo before launching the devcontainer:
-
-```
-# from the parent directory of your cudf clone
-git clone https://github.com/rapidsai/rapidsmpf.git
-```
-
 ## Launch a Dev Container
 
 To launch a devcontainer from VSCode, open the cuDF repo and select the "Reopen in Container" button in the bottom right:<br/><img src="https://user-images.githubusercontent.com/178183/221771999-97ab29d5-e718-4e5f-b32f-2cdd51bba25c.png"/>
diff --git a/.devcontainer/cuda12.9-conda/devcontainer.json b/.devcontainer/cuda12.9-conda/devcontainer.json
index f8581936a262..5326e64d8c6c 100644
--- a/.devcontainer/cuda12.9-conda/devcontainer.json
+++ b/.devcontainer/cuda12.9-conda/devcontainer.json
@@ -52,7 +52,7 @@
   "initializeCommand": [
     "/bin/bash",
     "-c",
-    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config,conda/pkgs,conda/${localWorkspaceFolderBasename}-cuda12.9-envs} ${localWorkspaceFolder}/../rapidsmpf"
+    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config,conda/pkgs,conda/${localWorkspaceFolderBasename}-cuda12.9-envs}"
   ],
   "postAttachCommand": [
     "/bin/bash",
@@ -66,8 +66,7 @@
     "source=${localWorkspaceFolder}/../.cache,target=/home/coder/.cache,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.config,target=/home/coder/.config,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.conda/pkgs,target=/home/coder/.conda/pkgs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../.conda/${localWorkspaceFolderBasename}-cuda12.9-envs,target=/home/coder/.conda/envs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../rapidsmpf,target=/home/coder/rapidsmpf,type=bind,consistency=consistent"
+    "source=${localWorkspaceFolder}/../.conda/${localWorkspaceFolderBasename}-cuda12.9-envs,target=/home/coder/.conda/envs,type=bind,consistency=consistent"
   ],
   "customizations": {
     "vscode": {
diff --git a/.devcontainer/cuda12.9-pip/devcontainer.json b/.devcontainer/cuda12.9-pip/devcontainer.json
index 03851a926c8c..914237875c0f 100644
--- a/.devcontainer/cuda12.9-pip/devcontainer.json
+++ b/.devcontainer/cuda12.9-pip/devcontainer.json
@@ -30,7 +30,7 @@
   "initializeCommand": [
     "/bin/bash",
     "-c",
-    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config/pip,local/share/${localWorkspaceFolderBasename}-cuda12.9-venvs} ${localWorkspaceFolder}/../rapidsmpf"
+    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config/pip,local/share/${localWorkspaceFolderBasename}-cuda12.9-venvs}"
   ],
   "postAttachCommand": [
     "/bin/bash",
@@ -43,8 +43,7 @@
     "source=${localWorkspaceFolder}/../.aws,target=/home/coder/.aws,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.cache,target=/home/coder/.cache,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.config,target=/home/coder/.config,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../.local/share/${localWorkspaceFolderBasename}-cuda12.9-venvs,target=/home/coder/.local/share/venvs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../rapidsmpf,target=/home/coder/rapidsmpf,type=bind,consistency=consistent"
+    "source=${localWorkspaceFolder}/../.local/share/${localWorkspaceFolderBasename}-cuda12.9-venvs,target=/home/coder/.local/share/venvs,type=bind,consistency=consistent"
   ],
   "customizations": {
     "vscode": {
diff --git a/.devcontainer/cuda13.2-conda/devcontainer.json b/.devcontainer/cuda13.2-conda/devcontainer.json
index 78a2f7f64963..a7987bd05dff 100644
--- a/.devcontainer/cuda13.2-conda/devcontainer.json
+++ b/.devcontainer/cuda13.2-conda/devcontainer.json
@@ -52,7 +52,7 @@
   "initializeCommand": [
     "/bin/bash",
     "-c",
-    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config,conda/pkgs,conda/${localWorkspaceFolderBasename}-cuda13.2-envs} ${localWorkspaceFolder}/../rapidsmpf"
+    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config,conda/pkgs,conda/${localWorkspaceFolderBasename}-cuda13.2-envs}"
   ],
   "postAttachCommand": [
     "/bin/bash",
@@ -66,8 +66,7 @@
     "source=${localWorkspaceFolder}/../.cache,target=/home/coder/.cache,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.config,target=/home/coder/.config,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.conda/pkgs,target=/home/coder/.conda/pkgs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../.conda/${localWorkspaceFolderBasename}-cuda13.2-envs,target=/home/coder/.conda/envs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../rapidsmpf,target=/home/coder/rapidsmpf,type=bind,consistency=consistent"
+    "source=${localWorkspaceFolder}/../.conda/${localWorkspaceFolderBasename}-cuda13.2-envs,target=/home/coder/.conda/envs,type=bind,consistency=consistent"
   ],
   "customizations": {
     "vscode": {
diff --git a/.devcontainer/cuda13.2-pip/devcontainer.json b/.devcontainer/cuda13.2-pip/devcontainer.json
index 7a283715b2e8..181ba40473cc 100644
--- a/.devcontainer/cuda13.2-pip/devcontainer.json
+++ b/.devcontainer/cuda13.2-pip/devcontainer.json
@@ -30,7 +30,7 @@
   "initializeCommand": [
     "/bin/bash",
     "-c",
-    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config/pip,local/share/${localWorkspaceFolderBasename}-cuda13.2-venvs} ${localWorkspaceFolder}/../rapidsmpf"
+    "mkdir -m 0755 -p ${localWorkspaceFolder}/../.{aws,cache,config/pip,local/share/${localWorkspaceFolderBasename}-cuda13.2-venvs}"
   ],
   "postAttachCommand": [
     "/bin/bash",
@@ -43,8 +43,7 @@
     "source=${localWorkspaceFolder}/../.aws,target=/home/coder/.aws,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.cache,target=/home/coder/.cache,type=bind,consistency=consistent",
     "source=${localWorkspaceFolder}/../.config,target=/home/coder/.config,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../.local/share/${localWorkspaceFolderBasename}-cuda13.2-venvs,target=/home/coder/.local/share/venvs,type=bind,consistency=consistent",
-    "source=${localWorkspaceFolder}/../rapidsmpf,target=/home/coder/rapidsmpf,type=bind,consistency=consistent"
+    "source=${localWorkspaceFolder}/../.local/share/${localWorkspaceFolderBasename}-cuda13.2-venvs,target=/home/coder/.local/share/venvs,type=bind,consistency=consistent"
   ],
   "customizations": {
     "vscode": {

From 70a8ad5269e31d44dc64aeb1748e3216d3ea96b6 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Mon, 8 Jun 2026 16:42:26 +0000
Subject: [PATCH 12/14] Support finding RAPIDS wheels from site-packages in
 pure CMake builds

---
 cpp/CMakeLists.txt | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
index 064549c401dd..a927dde70d33 100644
--- a/cpp/CMakeLists.txt
+++ b/cpp/CMakeLists.txt
@@ -290,6 +290,15 @@ endfunction()
 # * conda environment -----------------------------------------------------------------------------
 rapids_cmake_support_conda_env(conda_env MODIFY_PREFIX_PATH)
 
+# ##################################################################################################
+# * find C++ libraries from RAPIDS wheels in in site-packages --------------------------------------
+find_package(Python COMPONENTS Interpreter)
+if(Python_FOUND)
+  rapids_cython_find_prefix_paths("${Python_EXECUTABLE}" wheel_prefix_paths)
+  set_property(GLOBAL PROPERTY FIND_LIBRARY_USE_LIB64_PATHS TRUE)
+  list(APPEND CMAKE_PREFIX_PATH ${wheel_prefix_paths})
+endif()
+
 # ##################################################################################################
 # * compiler options ------------------------------------------------------------------------------
 set(CMAKE_MODULE_PATH ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Modules ${CMAKE_MODULE_PATH})

From 3a9b5016c94ae2257a66418f1b48575652385168 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Mon, 8 Jun 2026 17:07:38 +0000
Subject: [PATCH 13/14] Add include

---
 cpp/CMakeLists.txt | 1 +
 1 file changed, 1 insertion(+)

diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
index a927dde70d33..960c1c61eb43 100644
--- a/cpp/CMakeLists.txt
+++ b/cpp/CMakeLists.txt
@@ -294,6 +294,7 @@ rapids_cmake_support_conda_env(conda_env MODIFY_PREFIX_PATH)
 # * find C++ libraries from RAPIDS wheels in in site-packages --------------------------------------
 find_package(Python COMPONENTS Interpreter)
 if(Python_FOUND)
+  include(${rapids-cmake-dir}/cython-core/find_prefix_paths.cmake)
   rapids_cython_find_prefix_paths("${Python_EXECUTABLE}" wheel_prefix_paths)
   set_property(GLOBAL PROPERTY FIND_LIBRARY_USE_LIB64_PATHS TRUE)
   list(APPEND CMAKE_PREFIX_PATH ${wheel_prefix_paths})

From ca414c10b2dcdc99f2b3f139cbe10237a91bbd77 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Mon, 8 Jun 2026 18:02:59 -0700
Subject: [PATCH 14/14] Migrate from_cudf_packed_columns

---
 python/cudf_polars/cudf_polars/engine/spmd.py |  9 +-
 .../actor_graph/collectives/allgather.py      |  8 +-
 .../cudf_streaming/integrations/__init__.py   |  7 +-
 .../cudf_streaming/integrations/partition.pxd |  6 ++
 .../cudf_streaming/integrations/partition.pyi |  6 ++
 .../cudf_streaming/integrations/partition.pyx | 86 ++++++++++++++++++-
 .../cudf_streaming/tests/test_table_chunk.py  |  7 +-
 7 files changed, 117 insertions(+), 12 deletions(-)

diff --git a/python/cudf_polars/cudf_polars/engine/spmd.py b/python/cudf_polars/cudf_polars/engine/spmd.py
index e7905c845c56..96fe41de72b1 100644
--- a/python/cudf_polars/cudf_polars/engine/spmd.py
+++ b/python/cudf_polars/cudf_polars/engine/spmd.py
@@ -11,7 +11,10 @@
 from typing import TYPE_CHECKING, Any, cast
 
 import pylibcudf as plc
-from cudf_streaming.integrations.partition import unpack_and_concat
+from cudf_streaming.integrations.partition import (
+    packed_data_from_cudf_packed_columns,
+    unpack_and_concat,
+)
 from pylibcudf.contiguous_split import pack
 from rapidsmpf import bootstrap
 from rapidsmpf.coll import AllGather
@@ -19,7 +22,6 @@
     new_communicator as single_communicator,
 )
 from rapidsmpf.communicator.ucxx import barrier
-from rapidsmpf.memory.packed_data import PackedData
 from rapidsmpf.progress_thread import ProgressThread
 from rapidsmpf.rmm_resource_adaptor import RmmResourceAdaptor
 from rapidsmpf.statistics import Statistics
@@ -161,12 +163,11 @@ def allgather_polars_dataframe(
 
     plc_table = plc.Table.from_arrow(local_df, stream=stream)
 
-    packed_data = PackedData.from_cudf_packed_columns(
+    packed_data = packed_data_from_cudf_packed_columns(
         pack(plc_table, stream),
         stream,
         ctx.br(),
     )
-
     # Bulk AllGather: each rank contributes once (sequence_number=0)
     allgather = AllGather(comm, op_id, ctx.br())
     try:
diff --git a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
index 1772e9bdb41e..8939e50d8121 100644
--- a/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
+++ b/python/cudf_polars/cudf_polars/streaming/actor_graph/collectives/allgather.py
@@ -6,9 +6,11 @@
 
 from typing import TYPE_CHECKING
 
-from cudf_streaming.integrations.partition import unpack_and_concat
+from cudf_streaming.integrations.partition import (
+    packed_data_from_cudf_packed_columns,
+    unpack_and_concat,
+)
 from pylibcudf.contiguous_split import pack
-from rapidsmpf.memory.packed_data import PackedData
 from rapidsmpf.streaming.coll.allgather import AllGather
 
 if TYPE_CHECKING:
@@ -70,7 +72,7 @@ def insert(self, sequence_number: int, chunk: TableChunk) -> None:
                 sequence_number,
                 # TODO: Avoid unnecessary copies.
                 # See https://github.com/rapidsai/rapidsmpf/issues/933
-                PackedData.from_cudf_packed_columns(
+                packed_data_from_cudf_packed_columns(
                     pack(
                         chunk.table_view(),
                         chunk.stream,
diff --git a/python/cudf_streaming/cudf_streaming/integrations/__init__.py b/python/cudf_streaming/cudf_streaming/integrations/__init__.py
index 743c0b3ec454..37f3e8a1575c 100644
--- a/python/cudf_streaming/cudf_streaming/integrations/__init__.py
+++ b/python/cudf_streaming/cudf_streaming/integrations/__init__.py
@@ -2,6 +2,9 @@
 # SPDX-License-Identifier: Apache-2.0
 """Collection of cuDF specific functions."""
 
-from cudf_streaming.integrations.partition import unpack_and_concat
+from cudf_streaming.integrations.partition import (
+    packed_data_from_cudf_packed_columns,
+    unpack_and_concat,
+)
 
-__all__ = ["unpack_and_concat"]
+__all__ = ["packed_data_from_cudf_packed_columns", "unpack_and_concat"]
diff --git a/python/cudf_streaming/cudf_streaming/integrations/partition.pxd b/python/cudf_streaming/cudf_streaming/integrations/partition.pxd
index 774ccb338c10..9d674b1bd4c0 100644
--- a/python/cudf_streaming/cudf_streaming/integrations/partition.pxd
+++ b/python/cudf_streaming/cudf_streaming/integrations/partition.pxd
@@ -1,6 +1,7 @@
 # SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES.
 # SPDX-License-Identifier: Apache-2.0
 
+from pylibcudf.contiguous_split cimport PackedColumns
 from pylibcudf.table cimport Table
 from rmm.pylibrmm.stream cimport Stream
 
@@ -27,3 +28,8 @@ cpdef object unspill_partitions(
     BufferResource br,
     object allow_overbooking,
 )
+cpdef object packed_data_from_cudf_packed_columns(
+    PackedColumns packed_columns,
+    Stream stream,
+    BufferResource br,
+)
diff --git a/python/cudf_streaming/cudf_streaming/integrations/partition.pyi b/python/cudf_streaming/cudf_streaming/integrations/partition.pyi
index 1133c65ab1c9..efb30be09176 100644
--- a/python/cudf_streaming/cudf_streaming/integrations/partition.pyi
+++ b/python/cudf_streaming/cudf_streaming/integrations/partition.pyi
@@ -4,6 +4,7 @@ from __future__ import annotations
 
 from collections.abc import Iterable
 
+from pylibcudf.contiguous_split import PackedColumns
 from pylibcudf.table import Table
 
 from rapidsmpf.memory.buffer_resource import BufferResource
@@ -39,3 +40,8 @@ def unspill_partitions(
     br: BufferResource,
     allow_overbooking: bool,
 ) -> list[PackedData]: ...
+def packed_data_from_cudf_packed_columns(
+    packed_columns: PackedColumns,
+    stream: Stream,
+    br: BufferResource,
+) -> PackedData: ...
diff --git a/python/cudf_streaming/cudf_streaming/integrations/partition.pyx b/python/cudf_streaming/cudf_streaming/integrations/partition.pyx
index cc607d841934..f9fdc5b62203 100644
--- a/python/cudf_streaming/cudf_streaming/integrations/partition.pyx
+++ b/python/cudf_streaming/cudf_streaming/integrations/partition.pyx
@@ -4,16 +4,18 @@
 
 from cython.operator cimport dereference as deref
 from cython.operator cimport postincrement
-from libc.stdint cimport uint32_t
+from libc.stdint cimport uint8_t, uint32_t
 from libcpp.memory cimport make_unique, unique_ptr
 from libcpp.unordered_map cimport unordered_map
 from libcpp.utility cimport move
 from libcpp.vector cimport vector
+from pylibcudf.contiguous_split cimport PackedColumns
 from pylibcudf.libcudf.table.table cimport table as cpp_table
 from pylibcudf.libcudf.table.table_view cimport table_view
 from pylibcudf.libcudf.types cimport size_type
 from pylibcudf.table cimport Table
 from rmm.librmm.cuda_stream_view cimport cuda_stream_view
+from rmm.librmm.device_buffer cimport device_buffer
 from rmm.pylibrmm.stream cimport Stream
 
 from rapidsmpf._detail.exception_handling cimport ex_handler
@@ -316,6 +318,32 @@ cdef extern from "<cudf_streaming/integrations/partition.hpp>" nogil:
         ) except +ex_handler
 
 
+cdef extern from *:
+    """
+    #include <rapidsmpf/error.hpp>
+    #include <rapidsmpf/memory/packed_data.hpp>
+    #include <rapidsmpf/memory/buffer_resource.hpp>
+    #include <rmm/device_buffer.hpp>
+
+    std::unique_ptr<rapidsmpf::PackedData> cpp_packed_data_from_buffers(
+        std::unique_ptr<std::vector<std::uint8_t>> metadata,
+        std::unique_ptr<rmm::device_buffer> gpu_data,
+        rmm::cuda_stream_view stream,
+        rapidsmpf::BufferResource* br
+    ) {
+        return std::make_unique<rapidsmpf::PackedData>(
+            std::move(metadata), br->move(std::move(gpu_data), stream)
+        );
+    }
+    """
+    unique_ptr[cpp_PackedData] cpp_packed_data_from_buffers(
+        unique_ptr[vector[uint8_t]] metadata,
+        unique_ptr[device_buffer] gpu_data,
+        cuda_stream_view stream,
+        cpp_BufferResource* br,
+    ) except +ex_handler nogil
+
+
 cpdef object unspill_partitions(
     object partitions,
     BufferResource br,
@@ -369,3 +397,59 @@ cpdef object unspill_partitions(
             ab,
         )
     return packed_data_vector_to_list(move(_ret), br)
+
+
+cpdef object packed_data_from_cudf_packed_columns(
+    PackedColumns packed_columns,
+    Stream stream,
+    BufferResource br,
+):
+    """
+    Construct a PackedData from a pylibcudf PackedColumns.
+
+    Takes ownership of the metadata and GPU data from the PackedColumns
+    object, leaving it empty.
+
+    Parameters
+    ----------
+    packed_columns
+        Packed columns from ``pylibcudf.contiguous_split.pack()``.
+        Must not already be empty (already released).
+    stream
+        The CUDA stream on which the preceding ``pack()`` call was performed.
+        Must be the same stream to ensure correct memory ordering.
+    br
+        Buffer resource for memory management.
+
+    Returns
+    -------
+    A new PackedData instance owning the packed column data.
+
+    Raises
+    ------
+    ValueError
+        If the PackedColumns object is empty (already released).
+
+    See Also
+    --------
+    pylibcudf.contiguous_split.pack
+    cudf_streaming.integrations.partition.unpack_and_concat
+    """
+    if packed_columns is None or stream is None or br is None:
+        raise TypeError("Arguments must not be None")
+    cdef cuda_stream_view _stream = stream.view()
+    cdef cpp_BufferResource* _br = br.ptr()
+    cdef PackedData ret = PackedData.__new__(PackedData)
+    with nogil:
+        if not (packed_columns.c_obj != NULL and
+                deref(packed_columns.c_obj).metadata and
+                deref(packed_columns.c_obj).gpu_data):
+            raise ValueError("Cannot release empty PackedColumns")
+        ret.c_obj = cpp_packed_data_from_buffers(
+            move(deref(packed_columns.c_obj).metadata),
+            move(deref(packed_columns.c_obj).gpu_data),
+            _stream,
+            _br,
+        )
+    ret._br = br
+    return ret
diff --git a/python/cudf_streaming/cudf_streaming/tests/test_table_chunk.py b/python/cudf_streaming/cudf_streaming/tests/test_table_chunk.py
index f48a83e84131..6b55abd01f3c 100644
--- a/python/cudf_streaming/cudf_streaming/tests/test_table_chunk.py
+++ b/python/cudf_streaming/cudf_streaming/tests/test_table_chunk.py
@@ -9,6 +9,9 @@
 import pylibcudf as plc
 import pytest
 
+from cudf_streaming.integrations.partition import (
+    packed_data_from_cudf_packed_columns,
+)
 from cudf_streaming.streaming.table_chunk import (
     TableChunk,
     make_table_chunks_available_or_wait,
@@ -406,7 +409,7 @@ def test_shape_accessor(
     expected_shape = (expect.num_rows(), expect.num_columns())
 
     if from_pack:
-        pd = PackedData.from_cudf_packed_columns(
+        pd = packed_data_from_cudf_packed_columns(
             plc.contiguous_split.pack(expect, stream), stream, context.br()
         )
         device_chunk = TableChunk.from_packed_data(pd, br=context.br())
@@ -444,7 +447,7 @@ def test_into_packed_data(
 ) -> None:
     expect = random_table(1024)
     if from_pack:
-        pd = PackedData.from_cudf_packed_columns(
+        pd = packed_data_from_cudf_packed_columns(
             plc.contiguous_split.pack(expect, stream), stream, context.br()
         )
         chunk = TableChunk.from_packed_data(pd, br=context.br())