Merge branch 'master' into cuda-wheel-dep

.github/CHANGELOG.md

@@ -84,11 +84,17 @@
 
 ### Documentation
 
+* Add the exact tensor network to the Lightning-Tensor docs.
+  [(#1021)](https://github.com/PennyLaneAI/pennylane-lightning/pull/1021)
+
 * Update Lightning-Tensor installation docs and usage suggestions.
   [(#979)](https://github.com/PennyLaneAI/pennylane-lightning/pull/979)
 
 ### Bug fixes
 
+* Update CUDA version to 12.4 for GPU tests on CI.
+  [(#1023)](https://github.com/PennyLaneAI/pennylane-lightning/pull/1023)
+
 * Fix Lightning Kokkos `[[maybe_unused]]` and `exp2` errors with hipcc.
   [(#1018)](https://github.com/PennyLaneAI/pennylane-lightning/pull/1018)
 

.github/workflows/tests_gpu_cpp.yml

@@ -40,14 +40,15 @@ jobs:
     strategy:
       max-parallel: 1
       matrix:
-        cuda_version: ["12"]
+        cuda_version_maj: ["12"]
+        cuda_version_min: ["4"]
 
     steps:
       - name: Validate GPU version and installed compiler
         run: |
           source /etc/profile.d/modules.sh
           module use /opt/modules
-          module load cuda/${{ matrix.cuda_version }}
+          module load cuda/${{ matrix.cuda_version_maj }}.${{ matrix.cuda_version_min }}
           echo "${PATH}" >> $GITHUB_PATH
           echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}" >> $GITHUB_ENV
           nvcc --version

.github/workflows/tests_gpu_python.yml

@@ -40,14 +40,15 @@ jobs:
     strategy:
       max-parallel: 1
       matrix:
-        cuda_version: ["12"]
+        cuda_version_maj: ["12"]
+        cuda_version_min: ["4"]
 
     steps:
       - name: Validate GPU version and installed compiler
         run: |
           source /etc/profile.d/modules.sh
           module use /opt/modules
-          module load cuda/${{ matrix.cuda_version }}
+          module load cuda/${{ matrix.cuda_version_maj }}.${{ matrix.cuda_version_min }}
           echo "${PATH}" >> $GITHUB_PATH
           echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}" >> $GITHUB_ENV
           nvcc --version

.github/workflows/tests_lkcuda_cpp.yml

@@ -48,15 +48,17 @@ jobs:
       max-parallel: 1
       matrix:
         os: [ubuntu-22.04]
-        exec_model: ["CUDA"]
         kokkos_version: ["4.5.00"]
+        exec_model: ["CUDA"]
+        cuda_version_maj: ["12"]
+        cuda_version_min: ["4"]
 
     steps:
       - name: Validate GPU version and installed compiler
         run: |
           source /etc/profile.d/modules.sh
           module use /opt/modules
-          module load cuda/12.2
+          module load cuda/${{ matrix.cuda_version_maj }}.${{ matrix.cuda_version_min }}
           echo "${PATH}" >> $GITHUB_PATH
           echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}" >> $GITHUB_ENV
           nvcc --version
@@ -110,8 +112,10 @@ jobs:
       matrix:
         os: [ubuntu-22.04]
         pl_backend: ["lightning_kokkos"]
-        exec_model: ["CUDA"]
         kokkos_version: ["4.5.00"]
+        exec_model: ["CUDA"]
+        cuda_version_maj: ["12"]
+        cuda_version_min: ["4"]
 
     name: C++ Tests (${{ matrix.pl_backend }}, kokkos-${{ matrix.kokkos_version }}, model-${{ matrix.exec_model }})
     runs-on:
@@ -124,7 +128,7 @@ jobs:
         run: |
           source /etc/profile.d/modules.sh
           module use /opt/modules
-          module load cuda/12.2
+          module load cuda/${{ matrix.cuda_version_maj }}.${{ matrix.cuda_version_min }}
           echo "${PATH}" >> $GITHUB_PATH
           echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}" >> $GITHUB_ENV
           nvcc --version
@@ -190,7 +194,7 @@ jobs:
         run: |
           cd main
           cmake . -BBuild -G Ninja \
-            -DCMAKE_BUILD_TYPE=RelWithDebInfo \
+            -DCMAKE_BUILD_TYPE=Release \
             -DBUILD_TESTS=ON \
             -DENABLE_PYTHON=OFF \
             -DSCIPY_OPENBLAS=$(python -c "import scipy_openblas32; print(scipy_openblas32.get_lib_dir())") \

.github/workflows/tests_lkcuda_python.yml

@@ -53,15 +53,17 @@ jobs:
       max-parallel: 1
       matrix:
         os: [ubuntu-22.04]
-        exec_model: ["CUDA"]
         kokkos_version: ["4.5.00"]
+        exec_model: ["CUDA"]
+        cuda_version_maj: ["12"]
+        cuda_version_min: ["4"]
 
     steps:
       - name: Validate GPU version and installed compiler
         run: |
           source /etc/profile.d/modules.sh
           module use /opt/modules
-          module load cuda/12.2
+          module load cuda/${{ matrix.cuda_version_maj }}.${{ matrix.cuda_version_min }}
           echo "${PATH}" >> $GITHUB_PATH
           echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}" >> $GITHUB_ENV
           nvcc --version
@@ -115,8 +117,10 @@ jobs:
       matrix:
         os: [ubuntu-22.04]
         pl_backend: ["lightning_kokkos", "all"]
-        exec_model: ["CUDA"]
         kokkos_version: ["4.5.00"]
+        exec_model: ["CUDA"]
+        cuda_version_maj: ["12"]
+        cuda_version_min: ["4"]
 
     name: Python Tests (${{ matrix.pl_backend }}, kokkos-${{ matrix.kokkos_version }}, model-${{ matrix.exec_model }})
     runs-on:
@@ -129,7 +133,7 @@ jobs:
         run: |
           source /etc/profile.d/modules.sh
           module use /opt/modules
-          module load cuda/12.2
+          module load cuda/${{ matrix.cuda_version_maj }}.${{ matrix.cuda_version_min }}
           echo "${PATH}" >> $GITHUB_PATH
           echo "LD_LIBRARY_PATH=${LD_LIBRARY_PATH}" >> $GITHUB_ENV
           nvcc --version

doc/lightning_tensor/device.rst

@@ -1,35 +1,27 @@
 Lightning Tensor device
 =======================
 
-The ``lightning.tensor`` device is a tensor network simulator device. The device is built on top of the `cutensornet <https://docs.nvidia.com/cuda/cuquantum/latest/cutensornet/index.html>`__ from the NVIDIA cuQuantum SDK, enabling GPU-accelerated simulation of quantum tensor network evolution. This device is designed to simulate large-scale quantum circuits using tensor networks. For small circuits, state-vector simulator plugins may be more suitable.
+The ``lightning.tensor`` device is a tensor network simulator, supporting both the Matrix Product State (MPS) and Exact Tensor Network methods. The device is built on top of the `cutensornet <https://docs.nvidia.com/cuda/cuquantum/latest/cutensornet/index.html>`__ from the NVIDIA cuQuantum SDK, enabling GPU-accelerated simulation of quantum tensor network evolution. This device is designed to simulate large-scale quantum circuits using tensor networks. For small circuits, state-vector simulator plugins may be more suitable.
 
-A ``lightning.tensor`` device can be loaded simply using:
+The ``lightning.tensor`` device defaults to the Matrix Product State (MPS) method, and can be loaded using:
 
 .. code-block:: python
 
     import pennylane as qml
     dev = qml.device("lightning.tensor", wires=100)
 
-By default, the device represents the quantum state approximated as a Matrix Product State (MPS).
+By default, the device approximates the quantum state using an MPS.
 The default setup for the MPS tensor network approximation is:
     - ``max_bond_dim`` (maximum bond dimension) defaults to ``128`` .
     - ``cutoff`` (singular value truncation threshold) defaults to ``0`` .
     - ``cutoff_mode`` (singular value truncation mode) defaults to ``abs`` , considering the absolute values of the singular values; Alternatively, users can opt to set ``cutoff_mode`` to ``rel`` to consider the relative values of the singular values.
-Note that the ``cutensornet`` will automatically determine the reduced extent of the bond dimension based on the lowest among the multiple truncation cutoffs (``max_bond_dim``, ``cutoff-abs`` and ``cutoff-rel``). For more details on how the ``cutoff`` works, please check it out the `cuQuantum documentation <https://docs.nvidia.com/cuda/cuquantum/latest/cutensornet/api/types.html#cutensornettensorsvdconfigattributes-t>`__.
+Note that the ``cutensornet`` will automatically determine the reduced extent of the bond dimension based on the lowest among the multiple truncation cutoffs (``max_bond_dim``, ``cutoff-abs`` and ``cutoff-rel``). For more details on how the ``cutoff`` works, please check the `cuQuantum documentation <https://docs.nvidia.com/cuda/cuquantum/latest/cutensornet/api/types.html#cutensornettensorsvdconfigattributes-t>`__.
 
-The ``lightning.tensor`` device dispatches all operations to be performed on a CUDA-capable GPU of generation SM 7.0 (Volta)
-and greater. This device supports both exact and finite shots measurements. Currently, the supported differentiation methods are parameter-shift and finite-diff. Note that the MPS backend of lightning.tensor supports multi-wire gates via Matrix Product Operators (MPO).
-
-The ``lightning.tensor`` device is designed for expectation value calculations. Measurements of ``qml.probs()`` or ``qml.state()`` return dense vectors of dimension :math:`2^{n_\text{qubits}}`, so they should only be used for small systems.
-
-.. note:: ``qml.Hermitian`` is currently only supported for single wires. You can use ``qml.pauli_decompose`` on smaller matrices to obtain a compatible Pauli decomposition in the meantime.
-
-Users also have the flexibility to customize these parameters according to their specific needs with:
+Users also have the flexibility to customize MPS parameters according to their specific needs with:
 
 .. code-block:: python
     
     import pennylane as qml
-    import numpy as np
     
     num_qubits = 100
 
@@ -41,7 +33,21 @@ Users also have the flexibility to customize these parameters according to their
 
     dev = qml.device("lightning.tensor", wires=num_qubits, method="mps", **device_kwargs_mps)
 
-The ``lightning.tensor`` device allows users to get quantum circuit gradients using the ``parameter-shift`` method. This can be enabled at the PennyLane ``QNode`` level with:
+Users can also run the ``lightning.tensor`` device in the **Exact Tensor Network** mode by setting the ``method`` argument to ``"tn"``:
+
+.. code-block:: python
+
+    import pennylane as qml
+    dev = qml.device("lightning.tensor", wires=100, method="tn")
+
+The lightning.tensor device dispatches all operations to be performed on a CUDA-capable GPU of generation SM 7.0+ (Volta and later)
+and greater. This device supports both exact and finite shots measurements. Currently, the supported differentiation methods are parameter-shift and finite-diff. Note that the MPS backend of ``lightning.tensor`` supports multi-wire gates via Matrix Product Operators (MPO).
+
+The ``lightning.tensor`` device is designed for expectation value calculations. Measurements of :func:`~pennylane.probs` or :func:`~pennylane.state` return dense vectors of dimension :math:`2^{\text{n_qubits}}`, so they should only be used for small systems.
+
+.. note:: Currently only single-wire :class:`~pennylane.Hermitian` observables are supported. You can use :func:`~pennylane.pauli_decompose` on smaller matrices to obtain a compatible Pauli decomposition in the meantime.
+
+The ``lightning.tensor`` device allows users to get quantum circuit gradients using the :func:`~pennylane.gradients.param_shift` method. This can be enabled at the PennyLane :class:`~pennylane.QNode` level with:
 
 .. code-block:: python
 
@@ -59,7 +65,7 @@ Note that as ``lightning.tensor`` cannot be cleaned up like other state-vector d
 Operations and observables support
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-The "lightning.tensor" supports all gate operations supported by PennyLane.
+The ``lightning.tensor`` supports all gate operations supported by PennyLane, with the exception of :class:`~pennylane.StatePrep`, which is *not supported* by the *Exact Tensor Network* method. 
 
 **Supported operations:**
 
@@ -128,9 +134,9 @@ The "lightning.tensor" supports all gate operations supported by PennyLane.
 
 **Supported observables:**
 
-The ``lightning.tensor`` supports all observables supported by the Lightning state-vector simulators, besides ``qml.SparseHamiltonian``, ``qml.Projector`` and limited support to ``qml.Hamiltonian``, ``qml.Prod`` since ``lightning.tensor`` only supports 1-wire Hermitian observables.
+The ``lightning.tensor`` supports all observables supported by the Lightning state-vector simulators, besides :class:`~pennylane.SparseHamiltonian`, :class:`~pennylane.Projector` and limited support to :class:`~pennylane.ops.op_math.Sum`, :class:`~pennylane.ops.op_math.Prod` since ``lightning.tensor`` only supports 1-wire :class:`~pennylane.Hermitian` observables.
 
-Users can not create a ``Hamiltonian`` or ``Prod`` observable from ``Hamiltonian`` observables.
+Users cannot create a :class:`~pennylane.ops.op_math.Sum` observable or a :class:`~pennylane.ops.op_math.Prod` observable from :class:`~pennylane.ops.op_math.Sum` observables.
 
 
 
@@ -143,7 +149,6 @@ Users can not create a ``Hamiltonian`` or ``Prod`` observable from ``Hamiltonian
 
     ~pennylane.ops.op_math.Exp
     ~pennylane.Hadamard
-    ~pennylane.Hamiltonian
     ~pennylane.Hermitian
     ~pennylane.Identity
     ~pennylane.PauliX