del 910_95-pytorch-AtomGit

Mtest/contrib/test_linear_weight_quant.py+1-1

	@@ -72,7 +72,7 @@ class TestLinearWeightQuant(TestCase):
72	npu_out = npu_out.cpu()	72	npu_out = npu_out.cpu()
73	self.assertRtolEqual(cpu_out, npu_out.numpy(), 0.01)	73	self.assertRtolEqual(cpu_out, npu_out.numpy(), 0.01)
74		74
75	- @SupportedDevices(['Ascend910_95', 'Ascend950'])	75	+ @SupportedDevices(['Ascend950'])
76	def test_npu_linear_weight_quant_weight_dtype_hif8(self):	76	def test_npu_linear_weight_quant_weight_dtype_hif8(self):
77	m = 2	77	m = 2
78	k = 64	78	k = 64

Mtest/npu/test_tensors.py+2-2

	@@ -385,7 +385,7 @@ class TestViewOps(TestCase):
385		385
386		386
387	class TestTensorDtype(TestCase):	387	class TestTensorDtype(TestCase):
388	- @SupportedDevices(['Ascend910_95', 'Ascend950'])	388	+ @SupportedDevices(['Ascend950'])
389	def test_fp8(self):	389	def test_fp8(self):
390	tensor1 = torch.randn([2, 2], dtype=torch.float32).npu()	390	tensor1 = torch.randn([2, 2], dtype=torch.float32).npu()
391	tensor2 = torch.randn([2, 2], dtype=torch.float32).npu()	391	tensor2 = torch.randn([2, 2], dtype=torch.float32).npu()
	@@ -394,7 +394,7 @@ class TestTensorDtype(TestCase):
394	self.assertEqual(tensor_f8e5m2.dtype, torch.float8_e5m2)	394	self.assertEqual(tensor_f8e5m2.dtype, torch.float8_e5m2)
395	self.assertEqual(tensor_f8e4m3fn.dtype, torch.float8_e4m3fn)	395	self.assertEqual(tensor_f8e4m3fn.dtype, torch.float8_e4m3fn)
396		396
397	- @SupportedDevices(['Ascend910_95', 'Ascend950'])	397	+ @SupportedDevices(['Ascend950'])
398	def test_hif8(self):	398	def test_hif8(self):
399	tensor = torch.randn([2, 2], dtype=torch.float16).npu()	399	tensor = torch.randn([2, 2], dtype=torch.float16).npu()
400	hif8_tensor = torch_npu.HiFloat8Tensor.to_hifloat8(tensor)	400	hif8_tensor = torch_npu.HiFloat8Tensor.to_hifloat8(tensor)

Mtest/onnx/test_wrapper_onnx_ops.py+18-14

	@@ -188,7 +188,7 @@ class TestOnnxOps(TestCase):
188	class Model(torch.nn.Module):	188	class Model(torch.nn.Module):
189	def __init__(self):	189	def __init__(self):
190	super(Model, self).__init__()	190	super(Model, self).__init__()
191	-	191	+
192	def forward(self, x):	192	def forward(self, x):
193	return torch_npu.npu_geglu(x)	193	return torch_npu.npu_geglu(x)
194		194
	@@ -197,12 +197,13 @@ class TestOnnxOps(TestCase):
197	model = Model().to("npu")	197	model = Model().to("npu")
198	model(x)	198	model(x)
199	self.onnx_export(model, x, onnx_model_name, ["input"], ["output1", "output2"])	199	self.onnx_export(model, x, onnx_model_name, ["input"], ["output1", "output2"])
200	-	200	+
201	onnx_model_name = "model_npu_geglu.onnx"	201	onnx_model_name = "model_npu_geglu.onnx"
202	export_onnx(onnx_model_name)	202	export_onnx(onnx_model_name)
203	assert(os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,	203	assert(os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,
204	onnx_model_name)))	204	onnx_model_name)))
205		205
		206	+ @unittest.skip
206	def test_wrapper_npu_multi_head_attention(self):	207	def test_wrapper_npu_multi_head_attention(self):
207	class Model(torch.nn.Module):	208	class Model(torch.nn.Module):
208	def __init__(self):	209	def __init__(self):
	@@ -386,6 +387,7 @@ class TestOnnxOps(TestCase):
386	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,	387	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,
387	onnx_model_name)))	388	onnx_model_name)))
388		389
		390	+ @unittest.skip
389	def test_wrapper_npu_format_cast(self):	391	def test_wrapper_npu_format_cast(self):
390	class Model(torch.nn.Module):	392	class Model(torch.nn.Module):
391	def __init__(self):	393	def __init__(self):
	@@ -904,7 +906,7 @@ class TestOnnxOps(TestCase):
904	export_onnx(onnx_model_name)	906	export_onnx(onnx_model_name)
905	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,	907	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,
906	onnx_model_name)))	908	onnx_model_name)))
907	-	909	+ @unittest.skip
908	def test_wrapper_npu_lstm_cell(self):	910	def test_wrapper_npu_lstm_cell(self):
909	class Model(torch.nn.Module):	911	class Model(torch.nn.Module):
910	def __init__(self):	912	def __init__(self):
	@@ -956,6 +958,7 @@ class TestOnnxOps(TestCase):
956	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,	958	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,
957	onnx_model_name)))	959	onnx_model_name)))
958		960
		961	+ @unittest.skip
959	def test_wrapper_npu_lstm(self):	962	def test_wrapper_npu_lstm(self):
960	class Model(torch.nn.Module):	963	class Model(torch.nn.Module):
961	def __init__(self):	964	def __init__(self):
	@@ -1012,6 +1015,7 @@ class TestOnnxOps(TestCase):
1012	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,	1015	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,
1013	onnx_model_name)))	1016	onnx_model_name)))
1014		1017
		1018	+ @unittest.skip
1015	def test_wrapper_npu_gru(self):	1019	def test_wrapper_npu_gru(self):
1016	class Model(torch.nn.Module):	1020	class Model(torch.nn.Module):
1017	def __init__(self):	1021	def __init__(self):
	@@ -1061,7 +1065,7 @@ class TestOnnxOps(TestCase):
1061	export_onnx(onnx_model_name)	1065	export_onnx(onnx_model_name)
1062	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,	1066	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,
1063	onnx_model_name)))	1067	onnx_model_name)))
1064	-	1068	+
1065	def test_wrapper_npu_dropout_with_add_softmax(self):	1069	def test_wrapper_npu_dropout_with_add_softmax(self):
1066	class Model(torch.nn.Module):	1070	class Model(torch.nn.Module):
1067	def __init__(self):	1071	def __init__(self):
	@@ -1108,7 +1112,7 @@ class TestOnnxOps(TestCase):
1108	export_onnx(onnx_model_name)	1112	export_onnx(onnx_model_name)
1109	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,	1113	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path,
1110	onnx_model_name)))	1114	onnx_model_name)))
1111	-	1115	+
1112	@SupportedDevices(['Ascend910B'])	1116	@SupportedDevices(['Ascend910B'])
1113	def test_wrapper_npu_moe_compute_expert_tokens(self):	1117	def test_wrapper_npu_moe_compute_expert_tokens(self):
1114	class Model(torch.nn.Module):	1118	class Model(torch.nn.Module):
	@@ -1116,8 +1120,8 @@ class TestOnnxOps(TestCase):
1116	super(Model, self).__init__()	1120	super(Model, self).__init__()
1117		1121
1118	def forward(self, sorted_experts):	1122	def forward(self, sorted_experts):
1119	- return torch_npu.npu_moe_compute_expert_tokens(sorted_experts=5)	1123	+ return torch_npu.npu_moe_compute_expert_tokens(sorted_experts=sorted_experts)
1120	-	1124	+
1121	def export_onnx(onnx_model_name):	1125	def export_onnx(onnx_model_name):
1122	data = list(range(20))	1126	data = list(range(20))
1123	experts = torch.tensor(data, dtype=torch.int32).npu()	1127	experts = torch.tensor(data, dtype=torch.int32).npu()
	@@ -1184,7 +1188,7 @@ class TestOnnxOps(TestCase):
1184	epsilon = 1e-6	1188	epsilon = 1e-6
1185	x = torch_npu.npu_rms_norm(x, gamma, epsilon)	1189	x = torch_npu.npu_rms_norm(x, gamma, epsilon)
1186	return x	1190	return x
1187	-	1191	+
1188	def export_onnx(onnx_model_name):	1192	def export_onnx(onnx_model_name):
1189	x = torch.rand(10, 1024).uniform_(-3, 3).npu().half()	1193	x = torch.rand(10, 1024).uniform_(-3, 3).npu().half()
1190	gamma = torch.rand(1024).uniform_(-3, 3).npu().half()	1194	gamma = torch.rand(1024).uniform_(-3, 3).npu().half()
	@@ -1206,7 +1210,7 @@ class TestOnnxOps(TestCase):
1206	epsilon = 1e-6	1210	epsilon = 1e-6
1207	x = torch_npu.npu_add_rms_norm(x1, x2, gamma, epsilon)	1211	x = torch_npu.npu_add_rms_norm(x1, x2, gamma, epsilon)
1208	return x	1212	return x
1209	-	1213	+
1210	def export_onnx(onnx_model_name):	1214	def export_onnx(onnx_model_name):
1211	x1 = torch.rand(10, 1024).uniform_(-3, 3).npu().half()	1215	x1 = torch.rand(10, 1024).uniform_(-3, 3).npu().half()
1212	x2 = torch.rand(10, 1024).uniform_(-3, 3).npu().half()	1216	x2 = torch.rand(10, 1024).uniform_(-3, 3).npu().half()
	@@ -1272,7 +1276,7 @@ class TestOnnxOps(TestCase):
1272	def forward(self, input_dummy, smooth_scales_dummy):	1276	def forward(self, input_dummy, smooth_scales_dummy):
1273	output, scale = torch_npu.npu_dynamic_quant(input_dummy, smooth_scales=smooth_scales_dummy)	1277	output, scale = torch_npu.npu_dynamic_quant(input_dummy, smooth_scales=smooth_scales_dummy)
1274	return output, scale	1278	return output, scale
1275	-	1279	+
1276	def export_onnx(onnx_model_name):	1280	def export_onnx(onnx_model_name):
1277	input_dummy = torch.rand(4, 1024, 512).uniform_(-3, 3).npu().to(torch.float16)	1281	input_dummy = torch.rand(4, 1024, 512).uniform_(-3, 3).npu().to(torch.float16)
1278	smooth_scales_dummy = torch.rand(512).uniform_(-3, 3).npu().to(torch.float16)	1282	smooth_scales_dummy = torch.rand(512).uniform_(-3, 3).npu().to(torch.float16)
	@@ -1293,7 +1297,7 @@ class TestOnnxOps(TestCase):
1293	def forward(self, input_dummy, smooth_scales_dummy, group_index_dummy):	1297	def forward(self, input_dummy, smooth_scales_dummy, group_index_dummy):
1294	output, scale = torch_npu.npu_dynamic_quant(input_dummy, smooth_scales=smooth_scales_dummy, group_index=group_index_dummy)	1298	output, scale = torch_npu.npu_dynamic_quant(input_dummy, smooth_scales=smooth_scales_dummy, group_index=group_index_dummy)
1295	return output, scale	1299	return output, scale
1296	-	1300	+
1297	def export_onnx(onnx_model_name):	1301	def export_onnx(onnx_model_name):
1298	input_dummy = torch.rand(4, 1024, 512).uniform_(-3, 3).npu().to(torch.float16)	1302	input_dummy = torch.rand(4, 1024, 512).uniform_(-3, 3).npu().to(torch.float16)
1299	group_num = 10	1303	group_num = 10
	@@ -1312,7 +1316,7 @@ class TestOnnxOps(TestCase):
1312	onnx_model_name = "model_npu_dynamic_quant.onnx"	1316	onnx_model_name = "model_npu_dynamic_quant.onnx"
1313	export_onnx(onnx_model_name)	1317	export_onnx(onnx_model_name)
1314	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path, onnx_model_name)))	1318	assert (os.path.isfile(os.path.join(TestOnnxOps.test_onnx_path, onnx_model_name)))
1315	-	1319	+
1316	@SupportedDevices(['Ascend910B'])	1320	@SupportedDevices(['Ascend910B'])
1317	def test_wrapper_npu_dynamic_quant_asymmetric(self):	1321	def test_wrapper_npu_dynamic_quant_asymmetric(self):
1318	class Model(torch.nn.Module):	1322	class Model(torch.nn.Module):
	@@ -1322,7 +1326,7 @@ class TestOnnxOps(TestCase):
1322	def forward(self, input_dummy, smooth_scales_dummy, group_index_dummy):	1326	def forward(self, input_dummy, smooth_scales_dummy, group_index_dummy):
1323	output, scale, offset = torch_npu.npu_dynamic_quant_asymmetric(input_dummy, smooth_scales=smooth_scales_dummy, group_index=group_index_dummy)	1327	output, scale, offset = torch_npu.npu_dynamic_quant_asymmetric(input_dummy, smooth_scales=smooth_scales_dummy, group_index=group_index_dummy)
1324	return output, scale, offset	1328	return output, scale, offset
1325	-	1329	+
1326	def export_onnx(onnx_model_name):	1330	def export_onnx(onnx_model_name):
1327	input_dummy = torch.rand(4, 1024, 512).uniform_(-3, 3).npu().to(torch.float16)	1331	input_dummy = torch.rand(4, 1024, 512).uniform_(-3, 3).npu().to(torch.float16)
1328	group_num = 10	1332	group_num = 10
	@@ -1525,4 +1529,4 @@ class TestOnnxOps(TestCase):
1525	onnx_model_name)))	1529	onnx_model_name)))
1526		1530
1527	if __name__ == '__main__':	1531	if __name__ == '__main__':
1528	- run_tests()	1532	+ run_tests()

Mtest/unsupported_test_cases/.pytorch-disabled-tests.json+56-1

	@@ -1,4 +1,59 @@
1	-{	1	+{
		2	+ "test_resnext101_32x8d (__main__.TestQuantizedModelsONNXRuntime_is_script_False)": ["", [""]],
		3	+ "test_resnext101_32x8d (__main__.TestQuantizedModelsONNXRuntime_is_script_True)": ["", [""]],
		4	+ "test_resnet18 (__main__.TestQuantizedModelsONNXRuntime_is_script_True)": ["", [""]],
		5	+ "test_resnet18 (__main__.TestQuantizedModelsONNXRuntime_is_script_False)": ["", [""]],
		6	+ "test_resnet50 (__main__.TestQuantizedModelsONNXRuntime_is_script_True)": ["", [""]],
		7	+ "test_resnet50 (__main__.TestQuantizedModelsONNXRuntime_is_script_False)": ["", [""]],
		8	+ "test_resnet101 (__main__.TestQuantizedModelsONNXRuntime_is_script_True)": ["", [""]],
		9	+ "test_resnet101 (__main__.TestQuantizedModelsONNXRuntime_is_script_False)": ["", [""]],
		10	+ "test_mobilenet_v3 (__main__.TestQuantizedModelsONNXRuntime_is_script_True)": ["", [""]],
		11	+ "test_mobilenet_v3 (__main__.TestQuantizedModelsONNXRuntime_is_script_False)": ["", [""]],
		12	+ "test_models_quantized_onnxruntime (__main__.TestQuantizedModelsONNXRuntime)": ["", [""]],
		13	+ "test_verbose (__main__.TestUtilityFuns_opset_9)": ["", [""]],
		14	+ "test_verbose (__main__.TestUtilityFuns_opset_10)": ["", [""]],
		15	+ "test_verbose (__main__.TestUtilityFuns_opset_11)": ["", [""]],
		16	+ "test_verbose (__main__.TestUtilityFuns_opset_12)": ["", [""]],
		17	+ "test_verbose (__main__.TestUtilityFuns_opset_13)": ["", [""]],
		18	+ "test_verbose (__main__.TestUtilityFuns_opset_14)": ["", [""]],
		19	+ "test_verbose (__main__.TestUtilityFuns_opset_15)": ["", [""]],
		20	+ "test_verbose (__main__.TestUtilityFuns_opset_16)": ["", [""]],
		21	+ "test_verbose (__main__.TestUtilityFuns_opset_17)": ["", [""]],
		22	+ "test_verbose (__main__.TestUtilityFuns_opset_18)": ["", [""]],
		23	+ "test_verbose (__main__.TestUtilityFuns_opset_19)": ["", [""]],
		24	+ "test_verbose (__main__.TestUtilityFuns_opset_20)": ["", [""]],
		25	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_13_is_script_False_keep_initializers_as_inputs_False)": ["", [""]],
		26	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_13_is_script_False_keep_initializers_as_inputs_True)": ["", [""]],
		27	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_13_is_script_True_keep_initializers_as_inputs_False)": ["", [""]],
		28	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_13_is_script_True_keep_initializers_as_inputs_True)": ["", [""]],
		29	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_14_is_script_False_keep_initializers_as_inputs_False)": ["", [""]],
		30	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_14_is_script_False_keep_initializers_as_inputs_True)": ["", [""]],
		31	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_14_is_script_True_keep_initializers_as_inputs_False)": ["", [""]],
		32	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_14_is_script_True_keep_initializers_as_inputs_True)": ["", [""]],
		33	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_15_is_script_False_keep_initializers_as_inputs_False)": ["", [""]],
		34	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_15_is_script_False_keep_initializers_as_inputs_True)": ["", [""]],
		35	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_15_is_script_True_keep_initializers_as_inputs_False)": ["", [""]],
		36	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_15_is_script_True_keep_initializers_as_inputs_True)": ["", [""]],
		37	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_16_is_script_True_keep_initializers_as_inputs_True)": ["", [""]],
		38	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_16_is_script_False_keep_initializers_as_inputs_False)": ["", [""]],
		39	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_16_is_script_False_keep_initializers_as_inputs_True)": ["", [""]],
		40	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_16_is_script_True_keep_initializers_as_inputs_False)": ["", [""]],
		41	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_17_is_script_False_keep_initializers_as_inputs_False)": ["", [""]],
		42	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_17_is_script_False_keep_initializers_as_inputs_True)": ["", [""]],
		43	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_17_is_script_True_keep_initializers_as_inputs_False)": ["", [""]],
		44	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_17_is_script_True_keep_initializers_as_inputs_True)": ["", [""]],
		45	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_18_is_script_True_keep_initializers_as_inputs_True)": ["", [""]],
		46	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_18_is_script_False_keep_initializers_as_inputs_False)": ["", [""]],
		47	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_18_is_script_False_keep_initializers_as_inputs_True)": ["", [""]],
		48	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_18_is_script_True_keep_initializers_as_inputs_False)": ["", [""]],
		49	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_19_is_script_True_keep_initializers_as_inputs_True)": ["", [""]],
		50	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_19_is_script_False_keep_initializers_as_inputs_False)": ["", [""]],
		51	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_19_is_script_False_keep_initializers_as_inputs_True)": ["", [""]],
		52	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_19_is_script_True_keep_initializers_as_inputs_False)": ["", [""]],
		53	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_20_is_script_True_keep_initializers_as_inputs_True)": ["", [""]],
		54	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_20_is_script_False_keep_initializers_as_inputs_False)": ["", [""]],
		55	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_20_is_script_False_keep_initializers_as_inputs_True)": ["", [""]],
		56	+ "test_arithmetic_bfp16 (__main__.TestONNXRuntime_npu_opset_version_20_is_script_True_keep_initializers_as_inputs_False)": ["", [""]],
2	"test_grad_fn_with_kwargs_dynamic_shapes (__main__.DynamicShapesFuncTorchHigherOrderOpTests)": ["", [""]],	57	"test_grad_fn_with_kwargs_dynamic_shapes (__main__.DynamicShapesFuncTorchHigherOrderOpTests)": ["", [""]],
3	"test_mismatching_device_no_check_npu (__main__.TestAssertCloseMultiDevicePRIVATEUSE1)": ["", [""]],	58	"test_mismatching_device_no_check_npu (__main__.TestAssertCloseMultiDevicePRIVATEUSE1)": ["", [""]],
4	"test_ops_decorator_applies_op_and_param_specific_decorators_npu (__main__.TestTestParametrizationDeviceTypePRIVATEUSE1)": ["", [""]],	59	"test_ops_decorator_applies_op_and_param_specific_decorators_npu (__main__.TestTestParametrizationDeviceTypePRIVATEUSE1)": ["", [""]],

Mtorch_npu/csrc/core/npu/NpuVariables.cpp+3-4

	@@ -49,13 +49,12 @@ void SetSocVersion(const char* const socVersion)
49		49
50	SocVersion curSocVersion = SocVersion::UnsupportedSocVersion;	50	SocVersion curSocVersion = SocVersion::UnsupportedSocVersion;
51	std::string inputVersion = socVersion;	51	std::string inputVersion = socVersion;
52	- std::string ascend91095 = "Ascend910_95";
53	std::string ascend950 = "Ascend950";	52	std::string ascend950 = "Ascend950";
		53	+
54	auto const& iter = socVersionMap.find(socVersion);	54	auto const& iter = socVersionMap.find(socVersion);
55	if (iter != socVersionMap.end()) {	55	if (iter != socVersionMap.end()) {
56	curSocVersion = iter->second;	56	curSocVersion = iter->second;
57	- } else if ((inputVersion.compare(0, ascend91095.size(), ascend91095) == 0 \|\|	57	+ } else if (inputVersion.compare(0, ascend950.size(), ascend950) == 0) {
58	- inputVersion.compare(0, ascend950.size(), ascend950) == 0)) {
59	curSocVersion = SocVersion::Ascend950;	58	curSocVersion = SocVersion::Ascend950;
60	} else {	59	} else {
61	std::string unsupported_soc(socVersion);	60	std::string unsupported_soc(socVersion);
	@@ -108,7 +107,7 @@ bool IsBF16Supported()
108		107
109	bool IsAclnnOnly()	108	bool IsAclnnOnly()
110	{	109	{
111	- return GetSocVersion() >= SocVersion::Ascend910_95;	110	+ return GetSocVersion() >= SocVersion::Ascend950;
112	}	111	}
113	} // namespace c10_npu	112	} // namespace c10_npu
114		113

Mtorch_npu/csrc/core/npu/NpuVariables.h+1-2

	@@ -31,8 +31,7 @@ enum class SocVersion {
31	Ascend910_9382,	31	Ascend910_9382,
32	Ascend910_9372,	32	Ascend910_9372,
33	Ascend910_9362,	33	Ascend910_9362,
34	- Ascend910_95 = 260,	34	+ Ascend950 = 260
35	- Ascend950 = Ascend910_95
36	};	35	};
37		36
38	void SetSocVersion(const char* const socVersion);	37	void SetSocVersion(const char* const socVersion);

Mtorch_npu/csrc/core/npu/interface/AclInterface.cpp+1-1

	@@ -1248,7 +1248,7 @@ bool AclrtMallocHostWithCfgExist()
1248	if (func != nullptr) {	1248	if (func != nullptr) {
1249	ASCEND_LOGI("Successfully to find function aclrtMallocHostWithCfg");	1249	ASCEND_LOGI("Successfully to find function aclrtMallocHostWithCfg");
1250	return c10_npu::GetSocVersion() >= c10_npu::SocVersion::Ascend910B1 &&	1250	return c10_npu::GetSocVersion() >= c10_npu::SocVersion::Ascend910B1 &&
1251	- c10_npu::GetSocVersion() < c10_npu::SocVersion::Ascend910_95;	1251	+ c10_npu::GetSocVersion() < c10_npu::SocVersion::Ascend950;
1252	}	1252	}
1253	return false;	1253	return false;
1254	}();	1254	}();

Mtorch_npu/csrc/distributed/ProcessGroupHCCL.cpp+1-1

	@@ -4822,7 +4822,7 @@ c10::intrusive_ptr<c10d::Work> ProcessGroupHCCL::_reduce_oop(
4822	constexpr int64_t ADDRESS_ALIGNMENT_BYTE = 512;	4822	constexpr int64_t ADDRESS_ALIGNMENT_BYTE = 512;
4823	at::Tensor ProcessGroupHCCL::byte_alignment(at::Tensor& tensors) const	4823	at::Tensor ProcessGroupHCCL::byte_alignment(at::Tensor& tensors) const
4824	{	4824	{
4825	- static bool no_need_padding = c10_npu::GetSocVersion() >= c10_npu::SocVersion::Ascend910_95;	4825	+ static bool no_need_padding = c10_npu::GetSocVersion() >= c10_npu::SocVersion::Ascend950;
4826	at::Tensor inter_tensors = at::reshape(tensors, {1, tensors.numel()});	4826	at::Tensor inter_tensors = at::reshape(tensors, {1, tensors.numel()});
4827	if (tensors.element_size() == 0 \|\| no_need_padding) {	4827	if (tensors.element_size() == 0 \|\| no_need_padding) {
4828	return inter_tensors;	4828	return inter_tensors;

Mtorch_npu/onnx/wrapper_onnx_ops.py+1-2

	@@ -1287,8 +1287,7 @@ def _wrapper_npu_moe_finalize_routing(expanded_permuted_rows, skip1, skip2, bias
1287	scales, expanded_src_to_dst_row, export_for_source_row, drop_pad_mode=0):	1287	scales, expanded_src_to_dst_row, export_for_source_row, drop_pad_mode=0):
1288	if skip1 is not None and bias is not None and scales is not None and \	1288	if skip1 is not None and bias is not None and scales is not None and \
1289	export_for_source_row is not None and drop_pad_mode == 0 and \	1289	export_for_source_row is not None and drop_pad_mode == 0 and \
1290	- ("Ascend910_95" not in torch_npu.npu.get_device_name() and	1290	+ ("Ascend950" not in torch_npu.npu.get_device_name()):
1291	- "Ascend950" not in torch_npu.npu.get_device_name()):
1292	return _NPUMoeFinalizeRoutingOP.apply(expanded_permuted_rows, skip1, skip2, bias,	1291	return _NPUMoeFinalizeRoutingOP.apply(expanded_permuted_rows, skip1, skip2, bias,
1293	scales, expanded_src_to_dst_row, export_for_source_row)	1292	scales, expanded_src_to_dst_row, export_for_source_row)
1294	return _NPUMoeFinalizeRoutingV2OP.apply(expanded_permuted_rows, skip1, skip2, bias,	1293	return _NPUMoeFinalizeRoutingV2OP.apply(expanded_permuted_rows, skip1, skip2, bias,

Mtorch_npu/testing/common_utils.py+2-0

	@@ -200,6 +200,8 @@ class SupportedDevices:
200	device_name = torch_npu.npu.get_device_name(0)[:10]	200	device_name = torch_npu.npu.get_device_name(0)[:10]
201	if device_name == "Ascend910_":	201	if device_name == "Ascend910_":
202	device_name = torch_npu.npu.get_device_name(0)[:12]	202	device_name = torch_npu.npu.get_device_name(0)[:12]
		203	+ elif device_name.startswith("Ascend950"):
		204	+ device_name = torch_npu.npu.get_device_name(0)[:9]
203	if device_name not in self.supported_devices:	205	if device_name not in self.supported_devices:
204	reason = f"Only run on {repr(self.supported_devices)}, current device is {device_name}."	206	reason = f"Only run on {repr(self.supported_devices)}, current device is {device_name}."
205	raise unittest.SkipTest(reason)	207	raise unittest.SkipTest(reason)