Tensor str

oneflow/api/python/framework/tensor.cpp

@@ -278,7 +278,7 @@ Maybe<Tensor> NewTensor(py::args args, py::kwargs kwargs, const DType* desired_d
                     || py::isinstance<py::str>(device_kwarg));
 
     if (py::isinstance<py::str>(device_kwarg)) {
-      device = DeviceExportUtil::New(py::cast<std::string>(device_kwarg));
+      device = DeviceExportUtil::ParseAndNew(py::cast<std::string>(device_kwarg));
     } else {
       device = py::cast<Symbol<Device>>(device_kwarg);
     }

oneflow/core/functional/impl/array_functor.cpp

@@ -1177,6 +1177,7 @@ class TensorGetItemFunctor {
       step[i] = slice.step();
     }
     bool is_identity = [&]() {
+      if (target_shape.NumAxes() == 0) { return false; }
       for (int i = 0; i < ndims; ++i) {
         if (start[i] != 0 || end[i] != x->shape()->At(i) || step[i] != 1) { return false; }
       }

oneflow/core/functional/tensor_index.cpp

@@ -66,6 +66,7 @@ Maybe<void> PrepareSliceIndices(const TensorIndex& index, const Shape& shape,
     CHECK_LT_OR_RETURN(dim, ndims) << "Invalid index for tensor of dimension " << ndims;
     if (index_item.IsSlice()) {
       const auto& slice = index_item.slice();
+      CHECK_GT_OR_RETURN(slice.step(), 0) << "Step must be greater than zero.";
       int64_t step = std::min(slice.step(), shape.At(dim));
       int64_t end = std::min(slice.end(), shape.At(dim));
       int64_t start = std::min(slice.start(), shape.At(dim));

python/oneflow/compatible/single_client/nn/modules/activation.py

@@ -881,7 +881,7 @@ class Mish(Module):
 
         >>> out = mish(input)
         >>> out
-        tensor([0.8651, 1.944 , 2.9865], dtype=oneflow.float32)
+        tensor([0.8651, 1.9440, 2.9865], dtype=oneflow.float32)
     """
 
     def __init__(self, inplace: bool = False):

python/oneflow/framework/docstr/activation.py

@@ -68,7 +68,7 @@
 
         >>> out = flow.F.gelu(input)
         >>> out
-        tensor([-0.1543,  0.    ,  0.3457], dtype=oneflow.float32)
+        tensor([-0.1543,  0.0000,  0.3457], dtype=oneflow.float32)
 
     See    
     :class:`~oneflow.nn.GELU` for more details.
@@ -170,7 +170,7 @@
         >>> input = flow.Tensor(x) 
         >>> out = flow.F.softsign(input)
         >>> out
-        tensor([0.5   , 0.6667, 0.75  ], dtype=oneflow.float32)
+        tensor([0.5000, 0.6667, 0.7500], dtype=oneflow.float32)
 
     See :class:`~oneflow.nn.Softsign` for more details.
 
@@ -232,7 +232,7 @@
 
         >>> out = flow.F.mish(input)
         >>> out
-        tensor([0.8651, 1.944 , 2.9865], dtype=oneflow.float32)
+        tensor([0.8651, 1.9440, 2.9865], dtype=oneflow.float32)
 
     See :
     class:`~oneflow.nn.Mish` for more details.
@@ -361,7 +361,7 @@
         >>> input = flow.Tensor(x)
         >>> out = flow.nn.functional.elu(input, alpha=1.0)
         >>> out
-        tensor([-0.3935,  0.    ,  0.5   ], dtype=oneflow.float32)
+        tensor([-0.3935,  0.0000,  0.5000], dtype=oneflow.float32)
     """,
 )
 add_docstr(

python/oneflow/framework/docstr/pooling.py

@@ -39,7 +39,7 @@
         >>> arr = np.array([[[ 0.0558, -0.6875, -1.6544, -0.6226,  0.1018,  0.0502, -1.2538, 0.1491]]])
         >>> input = flow.Tensor(arr, dtype=flow.float32)
         >>> flow.nn.functional.adaptive_avg_pool1d(input, output_size=[4])
-        tensor([[[-0.3158, -1.1385,  0.076 , -0.5524]]], dtype=oneflow.float32)
+        tensor([[[-0.3158, -1.1385,  0.0760, -0.5524]]], dtype=oneflow.float32)
 
     """,
 )

python/oneflow/framework/tensor.py

@@ -325,6 +325,12 @@ def _get_device(self):
     raise NotImplementedError("get_device is only available for GPU tensor.")
 
 
+def _format(self, format_spec):
+    if self.dim() == 0:
+        return self.tolist().__format__(format_spec)
+    return object.__format__(self, format_spec)
+
+
 def RegisterMethods():
     Tensor.__mul__ = lambda self, other: self.mul(other)
     Tensor.__rmul__ = lambda self, other: self.mul(other)
@@ -359,6 +365,7 @@ def RegisterMethods():
     Tensor.__rtruediv__ = _rtruediv
     Tensor.__neg__ = _neg
     Tensor.__pow__ = _pow
+    Tensor.__format__ = _format
     Tensor.uniform_ = _uniform_
     Tensor.kaiming_uniform_ = _kaiming_uniform
     Tensor.kaiming_normal_ = _kaiming_normal