fix apis for Orange Pi

mindnlp/__init__.py

@@ -31,7 +31,10 @@
 import mindspore
 from mindspore import context
 from mindspore._c_expression import MSContext # pylint: disable=no-name-in-module, import-error
-
+try:
+    from mindspore._c_expression import disable_multi_thread
+except:
+    disable_multi_thread = None
 # for different ascend devices
 if platform.system().lower() == 'linux':
     SOC = MSContext.get_instance().get_ascend_soc_version()
@@ -41,6 +44,9 @@
     if SOC in ('ascend910', 'ascend310b'):
         context.set_context(ascend_config={"precision_mode": "allow_mix_precision"})
 
+    if SOC == 'ascend310b' and disable_multi_thread is not None:
+        disable_multi_thread()
+
 # set mindnlp.core to torch
 from .utils.torch_proxy import initialize_torch_proxy, setup_metadata_patch
 from .utils.safetensors_patch import setup_safetensors_patch

mindnlp/core/nn/functional.py

@@ -763,7 +763,7 @@ def batch_norm(input, running_mean, running_var, weight=None, bias=None, trainin
     if bias is None:
         bias = ops.zeros(input.shape[1])
 
-    if use_pyboost():
+    if use_pyboost() and not ON_ORANGE_PI:
         return mint.nn.functional.batch_norm(
             input,
             running_mean,
@@ -787,7 +787,7 @@ def batch_norm(input, running_mean, running_var, weight=None, bias=None, trainin
 
 has_conv1d = hasattr(mint.nn.functional, 'conv1d')
 def conv1d(input, weight, bias=None, stride=1, padding=0, dilation=1, groups=1):
-    if use_pyboost() and has_conv1d:
+    if use_pyboost() and has_conv1d and not ON_ORANGE_PI:
         return mint.nn.functional.conv1d(input, weight, bias, stride, padding, dilation, groups)
     pad_mode = 'pad'
     pad = padding
@@ -819,7 +819,7 @@ def conv1d(input, weight, bias=None, stride=1, padding=0, dilation=1, groups=1):
 
 
 def conv2d(input, weight, bias=None, stride=1, padding=0, dilation=1, groups=1):
-    if use_pyboost():
+    if use_pyboost() and not ON_ORANGE_PI:
         return mint.nn.functional.conv2d(input, weight, bias, stride, padding, dilation, groups)
 
     pad_mode = 'pad'
@@ -829,7 +829,7 @@ def conv2d(input, weight, bias=None, stride=1, padding=0, dilation=1, groups=1):
     return ops.conv2d(input, weight, bias=bias, stride=stride, pad_mode=pad_mode, padding=padding, dilation=dilation, groups=groups)
 
 def conv3d(input, weight, bias=None, stride=1, padding=0, dilation=1, groups=1):
-    if use_pyboost():
+    if use_pyboost() and not ON_ORANGE_PI:
         return mint.nn.functional.conv3d(input, weight, bias, stride, padding, dilation, groups)
 
     pad_mode = 'pad'
@@ -1014,7 +1014,7 @@ def max_pool1d(input, kernel_size, stride=None, padding=0, dilation=1, ceil_mode
 
 
 def group_norm(input, num_groups, weight=None, bias=None, eps=1e-5):
-    if use_pyboost():
+    if use_pyboost() and not ON_ORANGE_PI:
         return mint.nn.functional.group_norm(input, num_groups, weight, bias, eps)
 
     input_shape = input.shape
@@ -1027,7 +1027,11 @@ def group_norm(input, num_groups, weight=None, bias=None, eps=1e-5):
     affine_param_shape[1] = C
     affine_param_shape = tuple(affine_param_shape)
     if weight is not None and bias is not None:
-        out = bias.view(affine_param_shape).addcmul(out, weight.view(affine_param_shape), 1)
+        if not ON_ORANGE_PI:
+            out = bias.view(affine_param_shape).addcmul(out, weight.view(affine_param_shape), 1)
+        else:
+            out = core.addcmul(bias.view(affine_param_shape), out, weight.view(affine_param_shape), value=1)
+
     elif weight is not None:
         out = out.mul(weight.view(affine_param_shape))
     elif bias is not None:

mindnlp/core/ops/creation.py

@@ -150,7 +150,7 @@ def linspace(start, end, steps, *, dtype=None, **kwargs):
     start = start.item() if isinstance(start, mindspore.Tensor) else start
     end = end.item() if isinstance(end, mindspore.Tensor) else end
     steps = steps.item() if isinstance(steps, mindspore.Tensor) else steps
-    if use_pyboost() and has_linspace:
+    if use_pyboost() and has_linspace and not ON_ORANGE_PI:
         return mindspore.mint.linspace(start, end, steps, dtype=dtype)
     return ops.linspace(start, end, steps).to(dtype)
 

mindnlp/core/ops/inplace.py

@@ -11,7 +11,7 @@
     inplace_clamp_tensor_op, inplace_copy_op, inplace_index_add_op, inplace_erfinv_op
 
 from mindnlp import core
-from ..configs import use_pyboost
+from ..configs import use_pyboost, ON_ORANGE_PI
 from ._inner import assign
 
 generator_step_ = 12
@@ -230,7 +230,7 @@ def inplace_clamp(self, min=None, max=None):
     return self
 
 def inplace_erfinv(self):
-    if self.device.type == 'npu':
+    if self.device.type == 'npu' and not ON_ORANGE_PI:
         inplace_erfinv_op(self)
     else:
         self.data = core.erfinv(self)

mindnlp/core/ops/pointwise.py

@@ -60,7 +60,7 @@ def arccosh(input):
 
 
 def add(input, other, *, alpha=1, out=None):
-    if use_pyboost() and has_add:
+    if use_pyboost() and has_add and not ON_ORANGE_PI:
         return call_ms_func(mindspore.mint.add, input, other, alpha=alpha, out=out)
     if alpha != 1:
         other = mul(alpha, other)
@@ -335,9 +335,44 @@ def erfc(input, *, out=None):
 
 def erfinv(input, *, out=None):
     if use_pyboost() and has_erfinv:
+        if ON_ORANGE_PI:
+            return erfinv_torch(input)
         return call_ms_func(mindspore.mint.erfinv, input, out=out)
     return call_ms_func(ops.erfinv, input, out=out)
 
+def erfinv_torch(x):
+    """
+    使用有理函数近似实现erfinv，适用于PyTorch张量
+    """
+    # # 检查输入范围
+    # if core.any((x < -1) | (x > 1)):
+    #     raise ValueError("erfinv(x) is only defined for x in [-1, 1]")
+
+    # 处理边界情况
+    sign = core.where(x > 0, 1.0, -1.0)
+    x = core.abs(x)
+
+    # Cody的有理函数近似
+    mask = x <= 0.7
+    x_sq = x * x
+
+    # 对于x <= 0.7的情况
+    p1 = 0.426170613044 + x_sq * (-0.304570194263 + x_sq * 0.152645863430)
+    q1 = 1.0 + x_sq * (-0.733058978416 + x_sq * 0.546875000000)
+    result1 = x * (p1 / q1)
+
+    # 对于x > 0.7的情况
+    t = core.sqrt(-core.log((1.0 - x)/2.0))
+    p2 = -0.322232431088 + t * (-1.00002368368 + t * (-0.342242088547 + 
+         t * (-0.0204231210245 + t * (-0.0000453642210148))))
+    q2 = 0.460398842078 + t * (0.588581570495 + t * (0.531103462366 + 
+         t * (0.103537752850 + t * 0.0038560700634)))
+    result2 = p2 / q2
+
+    # 合并结果
+    result = core.where(mask, result1, result2)
+
+    return sign * result
 
 # exp
 has_exp = hasattr(mindspore.mint, "exp")