diff --git a/.github/workflows/build.yaml b/.github/workflows/build.yaml
index 9a055d1dc51..2962b4438fe 100644
--- a/.github/workflows/build.yaml
+++ b/.github/workflows/build.yaml
@@ -113,22 +113,3 @@ jobs:
       sha: ${{ inputs.sha }}
       date: ${{ inputs.date }}
       package-name: cugraph
-  wheel-build-cugraph-equivariant:
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-24.12
-    with:
-      build_type: ${{ inputs.build_type || 'branch' }}
-      branch: ${{ inputs.branch }}
-      sha: ${{ inputs.sha }}
-      date: ${{ inputs.date }}
-      script: ci/build_wheel_cugraph-equivariant.sh
-  wheel-publish-cugraph-equivariant:
-    needs: wheel-build-cugraph-equivariant
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@branch-24.12
-    with:
-      build_type: ${{ inputs.build_type || 'branch' }}
-      branch: ${{ inputs.branch }}
-      sha: ${{ inputs.sha }}
-      date: ${{ inputs.date }}
-      package-name: cugraph-equivariant
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index bac56fec9f5..e82342dfd94 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -25,8 +25,6 @@ jobs:
       - wheel-tests-pylibcugraph
       - wheel-build-cugraph
       - wheel-tests-cugraph
-      - wheel-build-cugraph-equivariant
-      - wheel-tests-cugraph-equivariant
       - devcontainer
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@branch-24.12
@@ -161,21 +159,6 @@ jobs:
     with:
       build_type: pull-request
       script: ci/test_wheel_cugraph.sh
-  wheel-build-cugraph-equivariant:
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-24.12
-    with:
-      build_type: pull-request
-      script: ci/build_wheel_cugraph-equivariant.sh
-  wheel-tests-cugraph-equivariant:
-    needs: [wheel-build-cugraph-equivariant, changed-files]
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@branch-24.12
-    if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python
-    with:
-      build_type: pull-request
-      script: ci/test_wheel_cugraph-equivariant.sh
-      matrix_filter: map(select(.ARCH == "amd64"))
   devcontainer:
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/build-in-devcontainer.yaml@branch-24.12
diff --git a/.github/workflows/test.yaml b/.github/workflows/test.yaml
index 245ac018cec..72f76904f4b 100644
--- a/.github/workflows/test.yaml
+++ b/.github/workflows/test.yaml
@@ -58,13 +58,3 @@ jobs:
       date: ${{ inputs.date }}
       sha: ${{ inputs.sha }}
       script: ci/test_wheel_cugraph.sh
-  wheel-tests-cugraph-equivariant:
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@branch-24.12
-    with:
-      build_type: nightly
-      branch: ${{ inputs.branch }}
-      date: ${{ inputs.date }}
-      sha: ${{ inputs.sha }}
-      script: ci/test_wheel_cugraph-equivariant.sh
-      matrix_filter: map(select(.ARCH == "amd64"))
diff --git a/build.sh b/build.sh
index 1ab98fe4378..398582c74c0 100755
--- a/build.sh
+++ b/build.sh
@@ -31,7 +31,6 @@ VALIDARGS="
    cugraph-service
    cugraph-pyg
    cugraph-dgl
-   cugraph-equivariant
    cpp-mgtests
    cpp-mtmgtests
    docs
@@ -59,7 +58,6 @@ HELP="$0 [<target> ...] [<flag> ...]
    cugraph-service            - build the cugraph-service_client and cugraph-service_server Python package
    cugraph-pyg                - build the cugraph-pyg Python package
    cugraph-dgl                - build the cugraph-dgl extensions for DGL
-   cugraph-equivariant        - build the cugraph-equivariant Python package
    cpp-mgtests                - build libcugraph and libcugraph_etl MG tests. Builds MPI communicator, adding MPI as a dependency.
    cpp-mtmgtests              - build libcugraph MTMG tests. Adds UCX as a dependency (temporary).
    docs                       - build the docs
@@ -209,8 +207,7 @@ if hasArg uninstall; then
     # FIXME: if multiple versions of these packages are installed, this only
     # removes the latest one and leaves the others installed. build.sh uninstall
     # can be run multiple times to remove all of them, but that is not obvious.
-    pip uninstall -y pylibcugraph cugraph cugraph-service-client cugraph-service-server \
-        cugraph-dgl cugraph-pyg cugraph-equivariant
+    pip uninstall -y pylibcugraph cugraph cugraph-service-client cugraph-service-server
 fi
 
 if hasArg clean; then
@@ -346,15 +343,6 @@ if hasArg cugraph-dgl || hasArg all; then
     fi
 fi
 
-# Build and install the cugraph-equivariant Python package
-if hasArg cugraph-equivariant || hasArg all; then
-    if hasArg --clean; then
-        cleanPythonDir ${REPODIR}/python/cugraph-equivariant
-    else
-        python ${PYTHON_ARGS_FOR_INSTALL} ${REPODIR}/python/cugraph-equivariant
-    fi
-fi
-
 # Build the docs
 if hasArg docs || hasArg all; then
     if [ ! -d ${LIBCUGRAPH_BUILD_DIR} ]; then
diff --git a/ci/build_python.sh b/ci/build_python.sh
index 9a8f1227488..713ae85920e 100755
--- a/ci/build_python.sh
+++ b/ci/build_python.sh
@@ -52,7 +52,6 @@ rapids-conda-retry mambabuild \
 
 rapids-conda-retry mambabuild \
   --no-test \
-  --channel "${RAPIDS_CONDA_BLD_OUTPUT_DIR}" \
-  conda/recipes/cugraph-equivariant
+  --channel "${RAPIDS_CONDA_BLD_OUTPUT_DIR}"
 
 rapids-upload-conda-to-s3 python
diff --git a/ci/build_wheel.sh b/ci/build_wheel.sh
index 3c89d63538c..c4faebdfdcd 100755
--- a/ci/build_wheel.sh
+++ b/ci/build_wheel.sh
@@ -30,26 +30,22 @@ python -m pip wheel \
 sccache --show-adv-stats
 
 # pure-python packages should be marked as pure, and not have auditwheel run on them.
-if [[ ${package_name} == "cugraph-equivariant" ]]; then
-    RAPIDS_PY_WHEEL_NAME="${package_name}_${RAPIDS_PY_CUDA_SUFFIX}" RAPIDS_PY_WHEEL_PURE="1" rapids-upload-wheels-to-s3 python dist
-else
-    case "${RAPIDS_CUDA_VERSION}" in
-        12.*)
-            EXCLUDE_ARGS=(
-                --exclude "libcublas.so.12"
-                --exclude "libcublasLt.so.12"
-                --exclude "libcurand.so.10"
-                --exclude "libcusolver.so.11"
-                --exclude "libcusparse.so.12"
-                --exclude "libnvJitLink.so.12"
-            )
-        ;;
-        11.*)
-            EXCLUDE_ARGS=()
-        ;;
-    esac
-
-    mkdir -p final_dist
-    python -m auditwheel repair -w final_dist "${EXCLUDE_ARGS[@]}" dist/*
-    RAPIDS_PY_WHEEL_NAME="${package_name}_${RAPIDS_PY_CUDA_SUFFIX}" rapids-upload-wheels-to-s3 python final_dist
-fi
+case "${RAPIDS_CUDA_VERSION}" in
+    12.*)
+        EXCLUDE_ARGS=(
+            --exclude "libcublas.so.12"
+            --exclude "libcublasLt.so.12"
+            --exclude "libcurand.so.10"
+            --exclude "libcusolver.so.11"
+            --exclude "libcusparse.so.12"
+            --exclude "libnvJitLink.so.12"
+        )
+    ;;
+    11.*)
+        EXCLUDE_ARGS=()
+    ;;
+esac
+
+mkdir -p final_dist
+python -m auditwheel repair -w final_dist "${EXCLUDE_ARGS[@]}" dist/*
+RAPIDS_PY_WHEEL_NAME="${package_name}_${RAPIDS_PY_CUDA_SUFFIX}" rapids-upload-wheels-to-s3 python final_dist
diff --git a/ci/run_cugraph_equivariant_pytests.sh b/ci/run_cugraph_equivariant_pytests.sh
deleted file mode 100755
index 5d5a5fb05c2..00000000000
--- a/ci/run_cugraph_equivariant_pytests.sh
+++ /dev/null
@@ -1,9 +0,0 @@
-#!/bin/bash
-# Copyright (c) 2024, NVIDIA CORPORATION.
-
-set -euo pipefail
-
-# Support invoking run_cugraph_equivariant_pytests.sh outside the script directory
-cd "$(dirname "$(realpath "${BASH_SOURCE[0]}")")"/../python/cugraph-equivariant/cugraph_equivariant
-
-pytest --cache-clear "$@" .
diff --git a/ci/test_python.sh b/ci/test_python.sh
index a3a177dcfc6..646b61805cc 100755
--- a/ci/test_python.sh
+++ b/ci/test_python.sh
@@ -100,42 +100,5 @@ rapids-logger "pytest cugraph-service (single GPU)"
   --cov-report=xml:"${RAPIDS_COVERAGE_DIR}/cugraph-service-coverage.xml" \
   --cov-report=term
 
-# test cugraph-equivariant
-if [[ "${RAPIDS_CUDA_VERSION}" == "11.8.0" ]]; then
-  if [[ "${RUNNER_ARCH}" != "ARM64" ]]; then
-    rapids-mamba-retry env create --yes -f env.yaml -n test_cugraph_equivariant
-    set +u
-    conda activate test_cugraph_equivariant
-    set -u
-    rapids-mamba-retry install \
-      --channel "${CPP_CHANNEL}" \
-      --channel "${PYTHON_CHANNEL}" \
-      --channel conda-forge \
-      --channel nvidia \
-      "cugraph-equivariant=${RAPIDS_VERSION_MAJOR_MINOR}.*"
-    pip install e3nn==0.5.1
-
-    rapids-print-env
-
-    rapids-logger "pytest cugraph-equivariant"
-    ./ci/run_cugraph_equivariant_pytests.sh \
-      --junitxml="${RAPIDS_TESTS_DIR}/junit-cugraph-equivariant.xml" \
-      --cov-config=../../.coveragerc \
-      --cov=cugraph_equivariant \
-      --cov-report=xml:"${RAPIDS_COVERAGE_DIR}/cugraph-equivariant-coverage.xml" \
-      --cov-report=term
-
-    # Reactivate the test environment back
-    set +u
-    conda deactivate
-    conda activate test
-    set -u
-  else
-    rapids-logger "skipping cugraph-equivariant pytest on ARM64"
-  fi
-else
-  rapids-logger "skipping cugraph-equivariant pytest on CUDA!=11.8"
-fi
-
 rapids-logger "Test script exiting with value: $EXITCODE"
 exit ${EXITCODE}
diff --git a/ci/test_wheel_cugraph-equivariant.sh b/ci/test_wheel_cugraph-equivariant.sh
deleted file mode 100755
index 3be1d578964..00000000000
--- a/ci/test_wheel_cugraph-equivariant.sh
+++ /dev/null
@@ -1,32 +0,0 @@
-#!/bin/bash
-# Copyright (c) 2024, NVIDIA CORPORATION.
-
-set -eoxu pipefail
-
-package_name="cugraph-equivariant"
-
-mkdir -p ./dist
-RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen ${RAPIDS_CUDA_VERSION})"
-
-# Download the cugraph-equivariant built in the previous step
-RAPIDS_PY_WHEEL_NAME="${package_name}_${RAPIDS_PY_CUDA_SUFFIX}" RAPIDS_PY_WHEEL_PURE="1" rapids-download-wheels-from-s3 ./dist
-
-# determine pytorch source
-PKG_CUDA_VER="$(echo ${CUDA_VERSION} | cut -d '.' -f1,2 | tr -d '.')"
-PKG_CUDA_VER_MAJOR=${PKG_CUDA_VER:0:2}
-if [[ "${PKG_CUDA_VER_MAJOR}" == "12" ]]; then
-  PYTORCH_CUDA_VER="121"
-else
-  PYTORCH_CUDA_VER=$PKG_CUDA_VER
-fi
-PYTORCH_URL="https://download.pytorch.org/whl/cu${PYTORCH_CUDA_VER}"
-
-# echo to expand wildcard before adding `[extra]` requires for pip
-python -m pip install \
-    -v \
-    --extra-index-url "${PYTORCH_URL}" \
-    "$(echo ./dist/cugraph_equivariant_${RAPIDS_PY_CUDA_SUFFIX}*.whl)[test]" \
-    'e3nn' \
-    'torch>=2.3.0,<2.4'
-
-python -m pytest python/cugraph-equivariant/cugraph_equivariant/tests
diff --git a/conda/recipes/cugraph-equivariant/build.sh b/conda/recipes/cugraph-equivariant/build.sh
deleted file mode 100644
index f0ff1688b55..00000000000
--- a/conda/recipes/cugraph-equivariant/build.sh
+++ /dev/null
@@ -1,7 +0,0 @@
-#!/usr/bin/env bash
-
-# Copyright (c) 2024, NVIDIA CORPORATION.
-
-# This assumes the script is executed from the root of the repo directory
-
-./build.sh cugraph-equivariant
diff --git a/conda/recipes/cugraph-equivariant/meta.yaml b/conda/recipes/cugraph-equivariant/meta.yaml
deleted file mode 100644
index 9dc9d51fa48..00000000000
--- a/conda/recipes/cugraph-equivariant/meta.yaml
+++ /dev/null
@@ -1,39 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-
-{% set version = environ['RAPIDS_PACKAGE_VERSION'].lstrip('v') + environ.get('VERSION_SUFFIX', '') %}
-{% set minor_version = version.split('.')[0] + '.' + version.split('.')[1] %}
-{% set py_version = environ['CONDA_PY'] %}
-{% set date_string = environ['RAPIDS_DATE_STRING'] %}
-
-package:
-  name: cugraph-equivariant
-  version: {{ version }}
-
-source:
-  path: ../../..
-
-build:
-  number: {{ GIT_DESCRIBE_NUMBER }}
-  build:
-      number: {{ GIT_DESCRIBE_NUMBER }}
-      string: py{{ py_version }}_{{ date_string }}_{{ GIT_DESCRIBE_HASH }}_{{ GIT_DESCRIBE_NUMBER }}
-
-requirements:
-  host:
-    - python
-    - rapids-build-backend>=0.3.1,<0.4.0.dev0
-    - setuptools>=61.0.0
-  run:
-    - pylibcugraphops ={{ minor_version }}
-    - python
-
-tests:
-  imports:
-    - cugraph_equivariant
-
-about:
-  home: https://rapids.ai/
-  dev_url: https://github.com/rapidsai/cugraph
-  license: Apache-2.0
-  license_file: ../../../LICENSE
-  summary: GPU-accelerated equivariant convolutional layers.
diff --git a/dependencies.yaml b/dependencies.yaml
index 208b3bedd63..6aa1b86aa60 100644
--- a/dependencies.yaml
+++ b/dependencies.yaml
@@ -202,30 +202,6 @@ files:
       - test_python_common
       - depends_on_pylibwholegraph
       - depends_on_pytorch
-  py_build_cugraph_equivariant:
-    output: pyproject
-    pyproject_dir: python/cugraph-equivariant
-    extras:
-      table: build-system
-    includes:
-      - python_build_rapids
-      - python_build_wheel
-  py_run_cugraph_equivariant:
-    output: pyproject
-    pyproject_dir: python/cugraph-equivariant
-    extras:
-      table: project
-    includes:
-        # Deprecate pylibcugraphops
-      - depends_on_pylibcugraphops
-  py_test_cugraph_equivariant:
-    output: pyproject
-    pyproject_dir: python/cugraph-equivariant
-    extras:
-      table: project.optional-dependencies
-      key: test
-    includes:
-      - test_python_common
   py_build_cugraph_service_client:
     output: pyproject
     pyproject_dir: python/cugraph-service/client
diff --git a/python/cugraph-equivariant/LICENSE b/python/cugraph-equivariant/LICENSE
deleted file mode 120000
index 30cff7403da..00000000000
--- a/python/cugraph-equivariant/LICENSE
+++ /dev/null
@@ -1 +0,0 @@
-../../LICENSE
\ No newline at end of file
diff --git a/python/cugraph-equivariant/README.md b/python/cugraph-equivariant/README.md
deleted file mode 100644
index d5de8852709..00000000000
--- a/python/cugraph-equivariant/README.md
+++ /dev/null
@@ -1,5 +0,0 @@
-# cugraph-equivariant
-
-## Description
-
-cugraph-equivariant library provides fast symmetry-preserving (equivariant) operations and convolutional layers, to accelerate the equivariant neural networks in drug discovery and other domains.
diff --git a/python/cugraph-equivariant/cugraph_equivariant/VERSION b/python/cugraph-equivariant/cugraph_equivariant/VERSION
deleted file mode 120000
index d62dc733efd..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/VERSION
+++ /dev/null
@@ -1 +0,0 @@
-../../../VERSION
\ No newline at end of file
diff --git a/python/cugraph-equivariant/cugraph_equivariant/__init__.py b/python/cugraph-equivariant/cugraph_equivariant/__init__.py
deleted file mode 100644
index 20507bd9329..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/__init__.py
+++ /dev/null
@@ -1,14 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from cugraph_equivariant._version import __git_commit__, __version__
diff --git a/python/cugraph-equivariant/cugraph_equivariant/_version.py b/python/cugraph-equivariant/cugraph_equivariant/_version.py
deleted file mode 100644
index 940ebac74f7..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/_version.py
+++ /dev/null
@@ -1,35 +0,0 @@
-# Copyright (c) 2023-2024, NVIDIA CORPORATION.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import importlib.resources
-
-# Read VERSION file from the module that is symlinked to VERSION file
-# in the root of the repo at build time or copied to the module at
-# installation. VERSION is a separate file that allows CI build-time scripts
-# to update version info (including commit hashes) without modifying
-# source files.
-__version__ = (
-    importlib.resources.files(__package__).joinpath("VERSION").read_text().strip()
-)
-try:
-    __git_commit__ = (
-        importlib.resources.files(__package__)
-        .joinpath("GIT_COMMIT")
-        .read_text()
-        .strip()
-    )
-except FileNotFoundError:
-    __git_commit__ = ""
-
-__all__ = ["__git_commit__", "__version__"]
diff --git a/python/cugraph-equivariant/cugraph_equivariant/nn/__init__.py b/python/cugraph-equivariant/cugraph_equivariant/nn/__init__.py
deleted file mode 100644
index 8f4d8de0042..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/nn/__init__.py
+++ /dev/null
@@ -1,21 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from .tensor_product_conv import FullyConnectedTensorProductConv
-
-DiffDockTensorProductConv = FullyConnectedTensorProductConv
-
-__all__ = [
-    "FullyConnectedTensorProductConv",
-    "DiffDockTensorProductConv",
-]
diff --git a/python/cugraph-equivariant/cugraph_equivariant/nn/tensor_product_conv.py b/python/cugraph-equivariant/cugraph_equivariant/nn/tensor_product_conv.py
deleted file mode 100644
index 923edbfc44a..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/nn/tensor_product_conv.py
+++ /dev/null
@@ -1,283 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from typing import Optional, Sequence, Union, NamedTuple
-
-import torch
-from torch import nn
-from e3nn import o3
-from e3nn.nn import BatchNorm
-
-from cugraph_equivariant.utils import scatter_reduce
-
-try:
-    from pylibcugraphops.pytorch.operators import FusedFullyConnectedTensorProduct
-except ImportError as exc:
-    raise RuntimeError(
-        "FullyConnectedTensorProductConv is no longer supported in "
-        "cugraph-equivariant starting from version 24.08. It will be migrated "
-        "to the new `cuequivariance` package. Please use 24.06 release for the "
-        "legacy interface."
-    ) from exc
-
-
-class Graph(NamedTuple):
-    edge_index: torch.Tensor
-    size: tuple[int, int]
-
-
-class FullyConnectedTensorProductConv(nn.Module):
-    r"""Message passing layer for tensor products in DiffDock-like architectures.
-    The left operand of tensor product is the spherical harmonic representation
-    of edge vector; the right operand consists of node features in irreps.
-
-    .. math::
-        \sum_{b \in \mathcal{N}_a} Y\left(\hat{r}_{a b}\right)
-        \otimes_{\psi_{a b}} \mathbf{h}_b
-
-    where the path weights :math:`\psi_{a b}` can be constructed from edge
-    embeddings and scalar features using an MLP:
-
-    .. math::
-        \psi_{a b} = \operatorname{MLP}
-        \left(e_{a b}, \mathbf{h}_a^0, \mathbf{h}_b^0\right)
-
-    Users have the option to either directly input the weights or provide the
-    MLP parameters and scalar features from edges and nodes.
-
-    Parameters
-    ----------
-    in_irreps : e3nn.o3.Irreps
-        Irreps for the input node features.
-
-    sh_irreps : e3nn.o3.Irreps
-        Irreps for the spherical harmonic representations of edge vectors.
-
-    out_irreps : e3nn.o3.Irreps
-        Irreps for the output.
-
-    batch_norm : bool, optional (default=True)
-        If true, batch normalization is applied.
-
-    mlp_channels : sequence of ints, optional (default=None)
-        A sequence of integers defining number of neurons in each layer in MLP
-        before the output layer. If `None`, no MLP will be added. The input layer
-        contains edge embeddings and node scalar features.
-
-    mlp_activation : nn.Module or sequence of nn.Module, optional (default=nn.GELU())
-        A sequence of functions to be applied in between linear layers in MLP,
-        e.g., `nn.Sequential(nn.ReLU(), nn.Dropout(0.4))`.
-
-    e3nn_compat_mode: bool, optional (default=False)
-        cugraph-ops and e3nn use different memory layout for Irreps-tensors.
-        The last (fastest moving) dimension is num_channels for cugraph-ops and
-        ir.dim for e3nn. When enabled, the input and output of this layer will
-        follow e3nn's memory layout.
-
-    Examples
-    --------
-    Case 1: MLP with the input layer having 6 channels and 2 hidden layers
-    having 16 channels. edge_emb.size(1) must match the size of the input layer: 6
-
-    >>> conv1 = FullyConnectedTensorProductConv(in_irreps, sh_irreps, out_irreps,
-    >>>     mlp_channels=[6, 16, 16], mlp_activation=nn.ReLU()).cuda()
-    >>> out = conv1(src_features, edge_sh, edge_emb, graph)
-
-    Case 2: If `edge_emb` is constructed by concatenating scalar features from
-    edges, sources and destinations, as in DiffDock, the layer can accept each
-    scalar component separately:
-
-    >>> conv2 = FullyConnectedTensorProductConv(in_irreps, sh_irreps, out_irreps,
-    >>>     mlp_channels=[6, 16, 16], mlp_activation=nn.ReLU()).cuda()
-    >>> out = conv2(src_features, edge_sh, edge_scalars, graph,
-    >>>     src_scalars=src_scalars, dst_scalars=dst_scalars)
-
-    This allows a smaller GEMM in the first MLP layer by performing GEMM on each
-    component before indexing. The first-layer weights are split into sections
-    for edges, sources and destinations, in that order.This is equivalent to
-
-    >>> src, dst = graph.edge_index
-    >>> edge_emb = torch.hstack((edge_scalars, src_scalars[src], dst_scalars[dst]))
-    >>> out = conv2(src_features, edge_sh, edge_emb, graph)
-
-    Case 3: No MLP, `edge_emb` will be directly used as the tensor product weights:
-
-    >>> conv3 = FullyConnectedTensorProductConv(in_irreps, sh_irreps, out_irreps,
-    >>>     mlp_channels=None).cuda()
-    >>> out = conv3(src_features, edge_sh, edge_emb, graph)
-
-    """
-
-    def __init__(
-        self,
-        in_irreps: o3.Irreps,
-        sh_irreps: o3.Irreps,
-        out_irreps: o3.Irreps,
-        batch_norm: bool = True,
-        mlp_channels: Optional[Sequence[int]] = None,
-        mlp_activation: Union[nn.Module, Sequence[nn.Module]] = nn.GELU(),
-        e3nn_compat_mode: bool = False,
-    ):
-        super().__init__()
-        self.in_irreps = in_irreps
-        self.out_irreps = out_irreps
-        self.sh_irreps = sh_irreps
-        self.e3nn_compat_mode = e3nn_compat_mode
-
-        self.tp = FusedFullyConnectedTensorProduct(
-            in_irreps, sh_irreps, out_irreps, e3nn_compat_mode=e3nn_compat_mode
-        )
-
-        self.batch_norm = BatchNorm(out_irreps) if batch_norm else None
-
-        if mlp_activation is None:
-            mlp_activation = []
-        elif hasattr(mlp_activation, "__len__") and hasattr(
-            mlp_activation, "__getitem__"
-        ):
-            mlp_activation = list(mlp_activation)
-        else:
-            mlp_activation = [mlp_activation]
-
-        if mlp_channels is not None:
-            dims = list(mlp_channels) + [self.tp.weight_numel]
-            mlp = []
-            for i in range(len(dims) - 1):
-                mlp.append(nn.Linear(dims[i], dims[i + 1]))
-                if i != len(dims) - 2:
-                    mlp.extend(mlp_activation)
-            self.mlp = nn.Sequential(*mlp)
-        else:
-            self.mlp = None
-
-    def forward(
-        self,
-        src_features: torch.Tensor,
-        edge_sh: torch.Tensor,
-        edge_emb: torch.Tensor,
-        graph: tuple[torch.Tensor, tuple[int, int]],
-        src_scalars: Optional[torch.Tensor] = None,
-        dst_scalars: Optional[torch.Tensor] = None,
-        reduce: str = "mean",
-        edge_envelope: Optional[torch.Tensor] = None,
-    ) -> torch.Tensor:
-        """Forward pass.
-
-        Parameters
-        ----------
-        src_features : torch.Tensor
-            Source node features.
-            Shape: (num_src_nodes, in_irreps.dim)
-
-        edge_sh : torch.Tensor
-            The spherical harmonic representations of the edge vectors.
-            Shape: (num_edges, sh_irreps.dim)
-
-        edge_emb: torch.Tensor
-            Edge embeddings that are fed into MLPs to generate tensor product weights.
-            Shape: (num_edges, dim), where `dim` should be:
-            - `tp.weight_numel` when the layer does not contain MLPs.
-            - num_edge_scalars, when scalar features from edges, sources and
-              destinations are passed in separately.
-
-        graph : tuple
-            A tuple that stores the graph information, with the first element being
-            the adjacency matrix in COO, and the second element being its shape:
-            (num_src_nodes, num_dst_nodes).
-
-        src_scalars: torch.Tensor, optional
-            Scalar features of source nodes. See examples for usage.
-            Shape: (num_src_nodes, num_src_scalars)
-
-        dst_scalars: torch.Tensor, optional
-            Scalar features of destination nodes. See examples for usage.
-            Shape: (num_dst_nodes, num_dst_scalars)
-
-        reduce : str, optional (default="mean")
-            Reduction operator. Choose between "mean" and "sum".
-
-        edge_envelope: torch.Tensor, optional
-            Typically used as attenuation factors to fade out messages coming
-            from nodes close to the cutoff distance used to create the graph.
-            This is important to make the model smooth to the changes in node's
-            coordinates.
-            Shape: (num_edges,)
-
-        Returns
-        -------
-        torch.Tensor
-            Output node features.
-            Shape: (num_dst_nodes, out_irreps.dim)
-        """
-        edge_emb_size = edge_emb.size(-1)
-        src_scalars_size = 0 if src_scalars is None else src_scalars.size(-1)
-        dst_scalars_size = 0 if dst_scalars is None else dst_scalars.size(-1)
-
-        if self.mlp is None:
-            if self.tp.weight_numel != edge_emb_size:
-                raise RuntimeError(
-                    f"When MLP is not present, edge_emb's last dimension must "
-                    f"equal tp.weight_numel (but got {edge_emb_size} and "
-                    f"{self.tp.weight_numel})"
-                )
-        else:
-            total_size = edge_emb_size + src_scalars_size + dst_scalars_size
-            if self.mlp[0].in_features != total_size:
-                raise RuntimeError(
-                    f"The size of MLP's input layer ({self.mlp[0].in_features}) "
-                    f"does not match the total number of scalar features from "
-                    f"edge_emb, src_scalars and dst_scalars ({total_size})"
-                )
-
-        if reduce not in ["mean", "sum"]:
-            raise RuntimeError(
-                f"reduce argument must be either 'mean' or 'sum', got {reduce}."
-            )
-
-        (src, dst), (num_src_nodes, num_dst_nodes) = graph
-
-        if self.mlp is not None:
-            if src_scalars is None and dst_scalars is None:
-                tp_weights = self.mlp(edge_emb)
-            else:
-                w_edge, w_src, w_dst = torch.split(
-                    self.mlp[0].weight,
-                    (edge_emb_size, src_scalars_size, dst_scalars_size),
-                    dim=-1,
-                )
-                tp_weights = edge_emb @ w_edge.T + self.mlp[0].bias
-
-                if src_scalars is not None:
-                    tp_weights += (src_scalars @ w_src.T)[src]
-
-                if dst_scalars is not None:
-                    tp_weights += (dst_scalars @ w_dst.T)[dst]
-
-                tp_weights = self.mlp[1:](tp_weights)
-        else:
-            tp_weights = edge_emb
-
-        out = self.tp(src_features[src], edge_sh, tp_weights)
-
-        if edge_envelope is not None:
-            out = out * edge_envelope.view(-1, 1)
-
-        dtype = out.dtype
-        out = scatter_reduce(
-            out.float(), dst, dim=0, dim_size=num_dst_nodes, reduce=reduce
-        ).to(dtype)
-
-        if self.batch_norm:
-            out = self.batch_norm(out)
-
-        return out
diff --git a/python/cugraph-equivariant/cugraph_equivariant/tests/conftest.py b/python/cugraph-equivariant/cugraph_equivariant/tests/conftest.py
deleted file mode 100644
index 806e03e6d76..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/tests/conftest.py
+++ /dev/null
@@ -1,39 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import pytest
-import torch
-
-
-@pytest.fixture
-def example_scatter_data():
-    src_feat = torch.Tensor([3, 1, 0, 1, 1, 2])
-    dst_indices = torch.Tensor([0, 1, 2, 2, 3, 1])
-
-    results = {
-        "sum": torch.Tensor([3.0, 3.0, 1.0, 1.0]),
-        "mean": torch.Tensor([3.0, 1.5, 0.5, 1.0]),
-        "prod": torch.Tensor([3.0, 2.0, 0.0, 1.0]),
-        "amax": torch.Tensor([3.0, 2.0, 1.0, 1.0]),
-        "amin": torch.Tensor([3.0, 1.0, 0.0, 1.0]),
-    }
-
-    return src_feat, dst_indices, results
-
-
-@pytest.fixture
-def empty_scatter_data():
-    src_feat = torch.empty((0, 41))
-    dst_indices = torch.empty((0,))
-
-    return src_feat, dst_indices
diff --git a/python/cugraph-equivariant/cugraph_equivariant/tests/pytest.ini b/python/cugraph-equivariant/cugraph_equivariant/tests/pytest.ini
deleted file mode 100644
index 7b0a9f29fb1..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/tests/pytest.ini
+++ /dev/null
@@ -1,4 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-
-[pytest]
-addopts = --tb=native
diff --git a/python/cugraph-equivariant/cugraph_equivariant/tests/test_scatter.py b/python/cugraph-equivariant/cugraph_equivariant/tests/test_scatter.py
deleted file mode 100644
index d28a32edcb1..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/tests/test_scatter.py
+++ /dev/null
@@ -1,40 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import pytest
-import torch
-from cugraph_equivariant.utils import scatter_reduce
-
-
-@pytest.mark.parametrize("reduce", ["sum", "mean", "prod", "amax", "amin"])
-def test_scatter_reduce(example_scatter_data, reduce):
-    device = torch.device("cuda")
-    src, index, out_true = example_scatter_data
-    src = src.to(device)
-    index = index.to(device)
-
-    out = scatter_reduce(src, index, dim=0, dim_size=None, reduce=reduce)
-
-    assert torch.allclose(out.cpu(), out_true[reduce])
-
-
-def test_scatter_reduce_empty(empty_scatter_data):
-    device = torch.device("cuda")
-    src, index = empty_scatter_data
-    src = src.to(device)
-    index = index.to(device)
-
-    out = scatter_reduce(src, index, dim=0, dim_size=None)
-
-    assert out.numel() == 0
-    assert out.size(1) == src.size(1)
diff --git a/python/cugraph-equivariant/cugraph_equivariant/tests/test_tensor_product_conv.py b/python/cugraph-equivariant/cugraph_equivariant/tests/test_tensor_product_conv.py
deleted file mode 100644
index ce325c47aa0..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/tests/test_tensor_product_conv.py
+++ /dev/null
@@ -1,136 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-import pytest
-
-try:
-    from cugraph_equivariant.nn import FullyConnectedTensorProductConv
-except RuntimeError:
-    pytest.skip(
-        "Migrated to cuequivariance package starting from 24.08.",
-        allow_module_level=True,
-    )
-
-import torch
-from torch import nn
-from e3nn import o3
-from cugraph_equivariant.nn.tensor_product_conv import Graph
-
-device = torch.device("cuda")
-
-
-def create_random_graph(
-    num_src_nodes,
-    num_dst_nodes,
-    num_edges,
-    dtype=None,
-    device=None,
-):
-    row = torch.randint(num_src_nodes, (num_edges,), dtype=dtype, device=device)
-    col = torch.randint(num_dst_nodes, (num_edges,), dtype=dtype, device=device)
-    edge_index = torch.stack([row, col], dim=0)
-
-    return Graph(edge_index, (num_src_nodes, num_dst_nodes))
-
-
-@pytest.mark.parametrize("dtype", [torch.float32, torch.float16, torch.bfloat16])
-@pytest.mark.parametrize("e3nn_compat_mode", [True, False])
-@pytest.mark.parametrize("batch_norm", [True, False])
-@pytest.mark.parametrize(
-    "mlp_channels, mlp_activation, scalar_sizes",
-    [
-        [(30, 8, 8), nn.Sequential(nn.Dropout(0.3), nn.ReLU()), (15, 15, 0)],
-        [(7,), nn.GELU(), (2, 3, 2)],
-        [None, None, None],
-    ],
-)
-def test_tensor_product_conv_equivariance(
-    mlp_channels, mlp_activation, scalar_sizes, batch_norm, e3nn_compat_mode, dtype
-):
-    torch.manual_seed(12345)
-    to_kwargs = {"device": device, "dtype": dtype}
-
-    in_irreps = o3.Irreps("10x0e + 10x1e")
-    out_irreps = o3.Irreps("20x0e + 10x1e")
-    sh_irreps = o3.Irreps.spherical_harmonics(lmax=2)
-
-    tp_conv = FullyConnectedTensorProductConv(
-        in_irreps=in_irreps,
-        sh_irreps=sh_irreps,
-        out_irreps=out_irreps,
-        mlp_channels=mlp_channels,
-        mlp_activation=mlp_activation,
-        batch_norm=batch_norm,
-        e3nn_compat_mode=e3nn_compat_mode,
-    ).to(**to_kwargs)
-
-    num_src_nodes, num_dst_nodes = 9, 7
-    num_edges = 40
-    graph = create_random_graph(num_src_nodes, num_dst_nodes, num_edges, device=device)
-
-    edge_sh = torch.randn(num_edges, sh_irreps.dim, **to_kwargs)
-    src_features = torch.randn(num_src_nodes, in_irreps.dim, **to_kwargs)
-
-    rot = o3.rand_matrix()
-    D_in = tp_conv.in_irreps.D_from_matrix(rot).to(**to_kwargs)
-    D_sh = tp_conv.sh_irreps.D_from_matrix(rot).to(**to_kwargs)
-    D_out = tp_conv.out_irreps.D_from_matrix(rot).to(**to_kwargs)
-
-    if mlp_channels is None:
-        edge_emb = torch.randn(num_edges, tp_conv.tp.weight_numel, **to_kwargs)
-        src_scalars = dst_scalars = None
-    else:
-        if scalar_sizes:
-            edge_emb = torch.randn(num_edges, scalar_sizes[0], **to_kwargs)
-            src_scalars = (
-                None
-                if scalar_sizes[1] == 0
-                else torch.randn(num_src_nodes, scalar_sizes[1], **to_kwargs)
-            )
-            dst_scalars = (
-                None
-                if scalar_sizes[2] == 0
-                else torch.randn(num_dst_nodes, scalar_sizes[2], **to_kwargs)
-            )
-        else:
-            edge_emb = torch.randn(num_edges, tp_conv.mlp[0].in_features, **to_kwargs)
-            src_scalars = dst_scalars = None
-
-    # rotate before
-    torch.manual_seed(12345)
-    out_before = tp_conv(
-        src_features=src_features @ D_in.T,
-        edge_sh=edge_sh @ D_sh.T,
-        edge_emb=edge_emb,
-        graph=graph,
-        src_scalars=src_scalars,
-        dst_scalars=dst_scalars,
-    )
-
-    # rotate after
-    torch.manual_seed(12345)
-    out_after = (
-        tp_conv(
-            src_features=src_features,
-            edge_sh=edge_sh,
-            edge_emb=edge_emb,
-            graph=graph,
-            src_scalars=src_scalars,
-            dst_scalars=dst_scalars,
-        )
-        @ D_out.T
-    )
-
-    atol = 1e-3 if dtype == torch.float32 else 1e-1
-    if e3nn_compat_mode:
-        assert torch.allclose(out_before, out_after, rtol=1e-4, atol=atol)
diff --git a/python/cugraph-equivariant/cugraph_equivariant/tests/test_version.py b/python/cugraph-equivariant/cugraph_equivariant/tests/test_version.py
deleted file mode 100644
index e8b484fe16c..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/tests/test_version.py
+++ /dev/null
@@ -1,12 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-
-import cugraph_equivariant
-
-
-def test_version_constants_are_populated():
-    # __git_commit__ will only be non-empty in a built distribution
-    assert isinstance(cugraph_equivariant.__git_commit__, str)
-
-    # __version__ should always be non-empty
-    assert isinstance(cugraph_equivariant.__version__, str)
-    assert len(cugraph_equivariant.__version__) > 0
diff --git a/python/cugraph-equivariant/cugraph_equivariant/utils/__init__.py b/python/cugraph-equivariant/cugraph_equivariant/utils/__init__.py
deleted file mode 100644
index b4acfe8d090..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/utils/__init__.py
+++ /dev/null
@@ -1,18 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from .scatter import scatter_reduce
-
-__all__ = [
-    "scatter_reduce",
-]
diff --git a/python/cugraph-equivariant/cugraph_equivariant/utils/scatter.py b/python/cugraph-equivariant/cugraph_equivariant/utils/scatter.py
deleted file mode 100644
index 909fbc99365..00000000000
--- a/python/cugraph-equivariant/cugraph_equivariant/utils/scatter.py
+++ /dev/null
@@ -1,42 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from typing import Optional
-
-import torch
-
-
-def broadcast(src: torch.Tensor, ref: torch.Tensor, dim: int) -> torch.Tensor:
-    size = ((1,) * dim) + (-1,) + ((1,) * (ref.dim() - dim - 1))
-    return src.view(size).expand_as(ref)
-
-
-def scatter_reduce(
-    src: torch.Tensor,
-    index: torch.Tensor,
-    dim: int = 0,
-    dim_size: Optional[int] = None,  # value of out.size(dim)
-    reduce: str = "sum",  # "sum", "prod", "mean", "amax", "amin"
-):
-    # scatter() expects index to be int64
-    index = broadcast(index, src, dim).to(torch.int64)
-
-    size = list(src.size())
-
-    if dim_size is not None:
-        size[dim] = dim_size
-    else:
-        size[dim] = 0 if index.numel() == 0 else int(index.max()) + 1
-
-    out = torch.zeros(size, dtype=src.dtype, device=src.device)
-    return out.scatter_reduce_(dim, index, src, reduce, include_self=False)
diff --git a/python/cugraph-equivariant/setup.py b/python/cugraph-equivariant/setup.py
deleted file mode 100644
index acd0df3f717..00000000000
--- a/python/cugraph-equivariant/setup.py
+++ /dev/null
@@ -1,20 +0,0 @@
-# Copyright (c) 2024, NVIDIA CORPORATION.
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-from setuptools import find_packages, setup
-
-if __name__ == "__main__":
-    packages = find_packages(include=["cugraph_equivariant*"])
-    setup(
-        package_data={key: ["VERSION"] for key in packages},
-    )