Merge branch 'beta' into cloud

Author: refraction-ray

CHANGELOG.md

@@ -4,7 +4,7 @@
 
 ### Added
 
-- Add `tc.about()` to print related software versions and configs.
+- Add `tc.about()` to print related software versions and configs
 
 - Torch support is updraded to 2.0, and now support native vmap and native functional grad, and thus `vvag`. Still jit support is conflict with these functional transformations and be turned off by default
 

docs/source/quickstart.rst

@@ -318,6 +318,7 @@ and the other part is implemented in `TensorCircuit package <modules.html#module
     'expm',
     'eye',
     'from_dlpack',
+    'g',
     'gather1d',
     'get_random_state',
     'gmres',
@@ -345,6 +346,7 @@ and the other part is implemented in `TensorCircuit package <modules.html#module
     'max',
     'mean',
     'min',
+    'minor',
     'mod',
     'multiply',
     'name',
@@ -353,6 +355,7 @@ and the other part is implemented in `TensorCircuit package <modules.html#module
     'one_hot',
     'onehot',
     'ones',
+    'optimizer',
     'outer_product',
     'pivot',
     'power',
@@ -419,6 +422,7 @@ and the other part is implemented in `TensorCircuit package <modules.html#module
     'vvag',
     'zeros']
 
+​
 
 Switch the Dtype
 --------------------
@@ -687,6 +691,20 @@ As we have mentioned in the backend section, the PyTorch backend may lack advanc
 
 For a GPU/CPU, torch/tensorflow, quantum/classical hybrid machine learning pipeline enabled by tensorcircuit, see `example script <https://github.com/tencent-quantum-lab/tensorcircuit/blob/master/examples/hybrid_gpu_pipeline.py>`__.
 
+There is also a more flexible torch interface that support static non-tensor inputs as keyword arguments, which can be utilized as below:
+
+.. code-block:: python
+
+    def f(a, i):
+    s = 0.
+    for _ in range(i):
+        s += a
+    return s
+
+    f_torch = tc.interfaces.torch_interface_kws(f)
+    f_torch(torch.ones([2]), i=3)
+
+
 We also provider wrapper of quantum function for torch module as :py:meth:`tensorcircuit.TorchLayer` alias to :py:meth:`tensorcircuit.torchnn.QuantumNet`.
 
 For ``TorchLayer``, ``use_interface=True`` is by default, which natively allow the quantum function defined on other tensorcircuit backends, such as jax or tf for speed consideration.

docs/source/sharpbits.rst

@@ -141,6 +141,53 @@ Similarly, conditional gate application must be takend carefully.
     # <tf.Tensor: shape=(2,), dtype=complex64, numpy=array([0.99999994+0.j, 0.        +0.j], dtype=complex64)>
 
 
+Tensor variables consistency
+-------------------------------------------------------
+
+
+All tensor variables' backend (tf vs jax vs ..), dtype (float vs complex), shape and device (cpu vs gpu) must be compatible/consistent.
+
+Inspect the backend, dtype, shape and device using the following codes.
+
+.. code-block:: python
+
+    for backend in ["numpy", "tensorflow", "jax", "pytorch"]:
+        with tc.runtime_backend(backend):
+            a = tc.backend.ones([2, 3])
+            print("tensor backend:", tc.interfaces.which_backend(a))
+            print("tensor dtype:", tc.backend.dtype(a))
+            print("tensor shape:", tc.backend.shape_tuple(a))
+            print("tensor device:", tc.backend.device(a))
+
+If the backend is inconsistent, one can convert the tensor backend via :py:meth:`tensorcircuit.interfaces.tensortrans.general_args_to_backend`.
+
+.. code-block:: python
+
+    for backend in ["numpy", "tensorflow", "jax", "pytorch"]:
+        with tc.runtime_backend(backend):
+            a = tc.backend.ones([2, 3])
+            print("tensor backend:", tc.interfaces.which_backend(a))
+            b = tc.interfaces.general_args_to_backend(a, target_backend="jax", enable_dlpack=False)
+            print("tensor backend:", tc.interfaces.which_backend(b))
+
+If the dtype is inconsistent, one can convert the tensor dtype using ``tc.backend.cast``.
+
+.. code-block:: python
+
+    for backend in ["numpy", "tensorflow", "jax", "pytorch"]:
+        with tc.runtime_backend(backend):
+            a = tc.backend.ones([2, 3])
+            print("tensor dtype:", tc.backend.dtype(a))
+            b = tc.backend.cast(a, dtype="float64")
+            print("tensor dtype:", tc.backend.dtype(b))
+
+Also note the jax issue on float64/complex128, see `jax gotcha <https://github.com/google/jax#current-gotchas>`_.
+
+If the shape is not consistent, one can convert the shape by ``tc.backend.reshape``.
+
+If the device is not consistent, one can move the tensor between devices by ``tc.backend.device_move``.
+
+
 AD Consistency
 ---------------------
 

docs/source/tutorial.rst

@@ -19,4 +19,5 @@ Jupyter Tutorials
     tutorials/optimization_and_expressibility.ipynb
     tutorials/vqex_mbl.ipynb
     tutorials/dqas.ipynb
-    tutorials/barren_plateaus.ipynb
+    tutorials/barren_plateaus.ipynb
+    tutorials/qaoa_portfolio_optimization.ipynb

docs/source/tutorials/qaoa.ipynb

@@ -28,7 +28,8 @@
 try:
     from . import quafu_provider
 except (ImportError, ModuleNotFoundError):
-    logger.warning("fail to load cloud provider module: quafu")
+    pass
+    # logger.warning("fail to load cloud provider module: quafu")
 
 package_name = "tensorcircuit"
 thismodule = sys.modules[__name__]

docs/source/tutorials/qaoa_portfolio_optimization.ipynb

@@ -135,7 +135,7 @@ def f(tensor, integer):
                 r += tensor
             return r
 
-        fnew = torch_interface_kws(f)
+        fnew = tc.interfaces.torch_interface_kws(f)
 
         print(fnew(torch.ones([2]), integer=3))
         print(fnew(torch.ones([2]), integer=4))