diff --git a/.cirun.yml b/.cirun.yml
index 2cff62d..12b9e0e 100644
--- a/.cirun.yml
+++ b/.cirun.yml
@@ -7,12 +7,11 @@ runners:
     instance_type: g4dn.xlarge  # Cheapest VM on AWS
     # Amazon AMI: Deep Learning AMI GPU CUDA 11.2.1 (Ubuntu 20.04) 20210625
     # for eu-west-1
-    machine_image: ami-07b138161bedd526d
+    machine_image: ami-00ac0c28c01352e53
     preemptible: false
-    workflow: .github/workflows/tests-gpu.yml
     count: 1
     labels:
-      - gpu
+      - "cirun-gpu-runner"
 
   - name: docs-gpu-runner
     cloud: aws
@@ -20,9 +19,8 @@ runners:
     instance_type: g4dn.xlarge  # Cheapest VM on AWS
     # Amazon AMI: Deep Learning AMI GPU CUDA 11.2.1 (Ubuntu 20.04) 20210625
     # for eu-west-1
-    machine_image: ami-07b138161bedd526d
+    machine_image: ami-00ac0c28c01352e53
     preemptible: false
-    workflow: .github/workflows/docs-gpu.yml
     count: 1
     labels:
-      - gpu
+      - "cirun-gpu-runner"
diff --git a/.github/workflows/tests-gpu.yml b/.github/workflows/tests-gpu.yml
index dbd2485..d2a1f66 100644
--- a/.github/workflows/tests-gpu.yml
+++ b/.github/workflows/tests-gpu.yml
@@ -58,7 +58,7 @@ jobs:
   cases:
     name: ${{ matrix.os }}, ${{ matrix.case-name }}
     needs: validate
-    runs-on: [self-hosted, gpu]
+    runs-on: [self-hosted, cirun-gpu-runner]
     strategy:
       fail-fast: false
       matrix:
diff --git a/pyproject.toml b/pyproject.toml
index da5fea8..df535c7 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,7 +1,7 @@
 [build-system]
 requires = [
     "setuptools>=36.6",
-    "packaging",
+    "packaging<=21.3",
     "wheel",
 ]
 build-backend = "setuptools.build_meta"
diff --git a/setup.cfg b/setup.cfg
index bb4a77f..fac5cf3 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -30,7 +30,7 @@ include_package_data = True
 install_requires =
     qutip>=5.0.0.dev0
 setup_requires =
-    packaging
+    packaging<=21.3
 
 [options.packages.find]
 where = src
diff --git a/src/qutip_cupy/__init__.py b/src/qutip_cupy/__init__.py
index ccdce21..c7eb476 100644
--- a/src/qutip_cupy/__init__.py
+++ b/src/qutip_cupy/__init__.py
@@ -80,14 +80,14 @@ def is_cupydense(data):
 data.norm.max.add_specialisations([(CuPyDense, cdf.max_cupydense)])
 data.norm.one.add_specialisations([(CuPyDense, cdf.one_cupydense)])
 
-data.inv.add_specialisations([(CuPyDense, linalg.inv_cupydense)])
+data.inv.add_specialisations([(CuPyDense, CuPyDense, linalg.inv_cupydense)])
 data.pow.add_specialisations([(CuPyDense, CuPyDense, cdf.pow_cupydense)])
 data.project.add_specialisations([(CuPyDense, CuPyDense, cdf.project_cupydense)])
 
 
 data.isherm.add_specialisations([(CuPyDense, cdf.isherm_cupydense)])
 
-data.inv.add_specialisations([(CuPyDense, linalg.inv_cupydense)])
+data.inv.add_specialisations([(CuPyDense, CuPyDense, linalg.inv_cupydense)])
 
 
 # We must register the functions to the data layer but do not want