fix bark ut and some numpy prims

mindnlp/core/_dtype.py

@@ -6,6 +6,8 @@
 
 from .configs import ON_A1
 
+bool_alias = bool
+
 if ON_A1:
     warnings.warn('MindSpore on GPU/910A do not support bfloat16, use float16 instead.')
     bfloat16 = float16
@@ -98,4 +100,8 @@ def __gt__(self, other):
     float16 : np.float16,
     float32 : np.float32,
     float64 : np.float64,
-}
+}
+
+py2dtype = {
+    bool_alias: bool
+}

mindnlp/core/_prims/ascend.py

@@ -137,6 +137,12 @@ def argmax_with_value(*args):
 
 __all__.append('argmax_with_value')
 
+def argmin_with_value(*args):
+    return pyboost_inner_prim.argmin_with_value_impl(*args)
+
+__all__.append('argmin_with_value')
+
+
 right_shift_op = ops.RightShift().set_device('Ascend')
 def right_shift(input, other):
     if isinstance(other, numbers.Number):

mindnlp/core/_prims/meta.py

@@ -262,3 +262,32 @@ def concat(tensors, dim):
     return core.Tensor(out)
 
 __all__.append('concat')
+
+def tril_ext(input, k):
+    return input
+
+__all__.append('tril_ext')
+
+def reshape(input, shape):
+    out = Tensor_(shape=tuple(shape), dtype=input.dtype)
+    return core.Tensor(out)
+
+__all__.append('reshape')
+
+def linalg_vector_norm(input, p, dim, keepdim, dtype):
+    input_shape = list(input.shape)
+    if isinstance(dim, int):
+        dim = (dim,)
+    for d in dim:
+        input_shape[d] = 1 if keepdim else 0
+
+    new_shape = []
+    for s in input_shape:
+        if s != 0:
+            new_shape.append(s)
+    if dtype is None:
+        dtype = input.dtype
+    out = Tensor_(shape=tuple(new_shape), dtype=dtype)
+    return core.Tensor(out)
+
+__all__.append('linalg_vector_norm')

mindnlp/core/_prims/numpy.py

@@ -61,6 +61,8 @@ def sub_ext(input, other, alpha):
     elif not isinstance(other, numbers.Number):
         other = other.numpy()
     out = np.subtract(input, other * alpha)
+    if not isinstance(out, np.ndarray):
+        out = np.array(out)
     return core.Tensor.from_numpy(out)
 
 __all__.append('sub_ext')
@@ -503,3 +505,9 @@ def less_equal(input, other):
     return core.Tensor.from_numpy(out)
 
 __all__.append('less_equal')
+
+def tril_ext(input, diagonal):
+    out = np.tril(input.numpy(), diagonal)
+    return core.Tensor.from_numpy(out)
+
+__all__.append('tril_ext')

mindnlp/core/ops/creation.py

@@ -23,7 +23,7 @@ def from_numpy(ndarray):
 
 # frombuffer
 def frombuffer(buffer, *, dtype, count=-1, offset=0, requires_grad=False):
-    arr = np.frombuffer(buffer=buffer, dtype=core.dtype_to_nptype(dtype), count=count, offset=offset)
+    arr = np.frombuffer(buffer=buffer, dtype=core.dtype2np[dtype], count=count, offset=offset)
     tensor = core.Tensor(arr)
     tensor.requires_grad_(requires_grad)
     return tensor
@@ -62,6 +62,8 @@ def zeros_like(input, *, dtype=None, layout=None, device=None, requires_grad=Fal
 def ones(*size, out=None, dtype=None, layout=None, device=None, requires_grad=False):
     if dtype is None:
         dtype = get_default_dtype()
+    if isinstance(dtype, type):
+        dtype = core.py2dtype[dtype]
     if device is None:
         device = get_device_in_context()
     if isinstance(size[0], (tuple, list)):

mindnlp/core/ops/other.py

@@ -866,7 +866,7 @@ def iinfo(dtype):
     return iinfo_dtype[dtype]
 
 def iinfo(dtype):
-    return np.iinfo(mindspore.dtype_to_nptype(dtype))
+    return np.iinfo(core.dtype2np[dtype])
 
 
 def contains(self, key):

mindnlp/core/ops/reduction.py

@@ -59,7 +59,7 @@ def min(input, dim=None, keepdim=False, *, out=None):
         return execute('min', input)
     if core.is_tensor(dim):
         return core.minimum(input, dim)
-    output = execute('argmin_ext', input, dim, keepdim)
+    output = execute('argmin_with_value', input, dim, keepdim)
     if out is None:
         return min_out(values=output[1], indices=output[0])