mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-07-27 20:02:15 +00:00
[python api] align api with other language bindings' treatment of explicit provider registrations. enforce use of providers param in python InferenceSession when execution providers other than default CPU are enabled. (#9712)
* remove default python ep registration. raise exception if providers are not explicitly set if there are available providers * temporarily disable exception * fix python tests * explicitly set CUDAProvider for python iobinding tests * explicitly set providers param for InferenceSession()) * onnxrt * raise ValueError if not explicitly set providers when creating InferenceSession * add required providers param * explicitly set providers * typo
This commit is contained in:
George Wu
GitHub
parent
517fff0a39
commit
1541784f6c
13 changed files with 98 additions and 80 deletions
|
|
@ -10,7 +10,7 @@ from onnx import helper
|
|||
from onnx import version
|
||||
from onnx.checker import check_model
|
||||
from onnx.backend.base import Backend
|
||||
from onnxruntime import InferenceSession, SessionOptions, get_device
|
||||
from onnxruntime import InferenceSession, SessionOptions, get_device, get_available_providers
|
||||
from onnxruntime.backend.backend_rep import OnnxRuntimeBackendRep
|
||||
import unittest
|
||||
import os
|
||||
|
|
@ -107,7 +107,7 @@ class OnnxRuntimeBackend(Backend):
|
|||
for k, v in kwargs.items():
|
||||
if hasattr(options, k):
|
||||
setattr(options, k, v)
|
||||
inf = InferenceSession(model, options)
|
||||
inf = InferenceSession(model, sess_options=options, providers=get_available_providers())
|
||||
# backend API is primarily used for ONNX test/validation. As such, we should disable session.run() fallback
|
||||
# which may hide test failures.
|
||||
inf.disable_fallback()
|
||||
|
|
|
|||
|
|
@ -356,13 +356,11 @@ class InferenceSession(Session):
|
|||
providers, provider_options = check_and_normalize_provider_args(providers,
|
||||
provider_options,
|
||||
available_providers)
|
||||
|
||||
if providers == [] and len(available_providers) > 1:
|
||||
warnings.warn("Deprecation warning. This ORT build has {} enabled. ".format(available_providers) +
|
||||
"The next release (ORT 1.10) will require explicitly setting the providers parameter " +
|
||||
"(as opposed to the current behavior of providers getting set/registered by default " +
|
||||
"based on the build flags) when instantiating InferenceSession."
|
||||
"For example, onnxruntime.InferenceSession(..., providers=[\"CUDAExecutionProvider\"], ...)")
|
||||
raise ValueError("This ORT build has {} enabled. ".format(available_providers) +
|
||||
"Since ORT 1.9, you are required to explicitly set " +
|
||||
"the providers parameter when instantiating InferenceSession. For example, "
|
||||
"onnxruntime.InferenceSession(..., providers={}, ...)".format(available_providers))
|
||||
|
||||
session_options = self._sess_options if self._sess_options else C.get_default_session_options()
|
||||
if self._model_path:
|
||||
|
|
|
|||
|
|
@ -783,12 +783,7 @@ void InitializeSession(InferenceSession* sess,
|
|||
ProviderOptionsMap provider_options_map;
|
||||
GenerateProviderOptionsMap(provider_types, provider_options, provider_options_map);
|
||||
|
||||
if (provider_types.empty()) {
|
||||
// use default registration priority.
|
||||
ep_registration_fn(sess, GetAllExecutionProviderNames(), provider_options_map);
|
||||
} else {
|
||||
ep_registration_fn(sess, provider_types, provider_options_map);
|
||||
}
|
||||
ep_registration_fn(sess, provider_types, provider_options_map);
|
||||
|
||||
#if !defined(ORT_MINIMAL_BUILD)
|
||||
if (!disabled_optimizer_names.empty()) {
|
||||
|
|
|
|||
|
|
@ -28,7 +28,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
so.log_verbosity_level = 1
|
||||
so.logid = "TestModelSerialization"
|
||||
so.optimized_model_filepath = "./PythonApiTestOptimizedModel.onnx"
|
||||
onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so)
|
||||
onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so, providers=['CPUExecutionProvider'])
|
||||
self.assertTrue(os.path.isfile(so.optimized_model_filepath))
|
||||
except Fail as onnxruntime_error:
|
||||
if str(onnxruntime_error) == "[ONNXRuntimeError] : 1 : FAIL : Unable to serialize model as it contains" \
|
||||
|
|
@ -42,7 +42,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
# get_all_providers() returns the default EP order from highest to lowest.
|
||||
# CPUExecutionProvider should always be last.
|
||||
self.assertTrue('CPUExecutionProvider' == onnxrt.get_all_providers()[-1])
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
self.assertTrue('CPUExecutionProvider' in sess.get_providers())
|
||||
|
||||
def testEnablingAndDisablingTelemetry(self):
|
||||
|
|
@ -54,7 +54,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
|
||||
def testSetProviders(self):
|
||||
if 'CUDAExecutionProvider' in onnxrt.get_available_providers():
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=['CUDAExecutionProvider'])
|
||||
# confirm that CUDA Provider is in list of registered providers.
|
||||
self.assertTrue('CUDAExecutionProvider' in sess.get_providers())
|
||||
# reset the session and register only CPU Provider.
|
||||
|
|
@ -64,7 +64,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
|
||||
def testSetProvidersWithOptions(self):
|
||||
if 'TensorrtExecutionProvider' in onnxrt.get_available_providers():
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=['TensorrtExecutionProvider'])
|
||||
self.assertIn('TensorrtExecutionProvider', sess.get_providers())
|
||||
|
||||
options = sess.get_provider_options()
|
||||
|
|
@ -127,7 +127,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
import ctypes
|
||||
CUDA_SUCCESS = 0
|
||||
def runBaseTest1():
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=['CUDAExecutionProvider'])
|
||||
self.assertTrue('CUDAExecutionProvider' in sess.get_providers())
|
||||
|
||||
option1 = {'device_id': 0}
|
||||
|
|
@ -140,7 +140,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
self.assertEqual(['CUDAExecutionProvider', 'CPUExecutionProvider'], sess.get_providers())
|
||||
|
||||
def runBaseTest2():
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=['CUDAExecutionProvider'])
|
||||
self.assertIn('CUDAExecutionProvider', sess.get_providers())
|
||||
|
||||
# test get/set of "gpu_mem_limit" configuration.
|
||||
|
|
@ -237,7 +237,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
result = ctypes.c_int()
|
||||
error_str = ctypes.c_char_p()
|
||||
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=['CPUExecutionProvider'])
|
||||
option = {'device_id': i}
|
||||
sess.set_providers(['CUDAExecutionProvider'], [option])
|
||||
self.assertEqual(['CUDAExecutionProvider', 'CPUExecutionProvider'], sess.get_providers())
|
||||
|
|
@ -258,7 +258,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
for i in range(num_device):
|
||||
setDeviceIdTest(i)
|
||||
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=['CPUExecutionProvider'])
|
||||
|
||||
# configure session with invalid option values and that should fail
|
||||
with self.assertRaises(RuntimeError):
|
||||
|
|
@ -291,7 +291,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
|
||||
def testInvalidSetProviders(self):
|
||||
with self.assertRaises(RuntimeError) as context:
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=['CPUExecutionProvider'])
|
||||
sess.set_providers(['InvalidProvider'])
|
||||
self.assertTrue('Unknown Provider Type: InvalidProvider' in str(context.exception))
|
||||
|
||||
|
|
@ -302,7 +302,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
self.assertEqual(['CPUExecutionProvider'], sess.get_providers())
|
||||
|
||||
def testRunModel(self):
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32)
|
||||
input_name = sess.get_inputs()[0].name
|
||||
self.assertEqual(input_name, "X")
|
||||
|
|
@ -319,7 +319,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
def testRunModelFromBytes(self):
|
||||
with open(get_name("mul_1.onnx"), "rb") as f:
|
||||
content = f.read()
|
||||
sess = onnxrt.InferenceSession(content)
|
||||
sess = onnxrt.InferenceSession(content, providers=onnxrt.get_available_providers())
|
||||
x = np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32)
|
||||
input_name = sess.get_inputs()[0].name
|
||||
self.assertEqual(input_name, "X")
|
||||
|
|
@ -334,7 +334,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_allclose(output_expected, res[0], rtol=1e-05, atol=1e-08)
|
||||
|
||||
def testRunModel2(self):
|
||||
sess = onnxrt.InferenceSession(get_name("matmul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("matmul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32)
|
||||
input_name = sess.get_inputs()[0].name
|
||||
self.assertEqual(input_name, "X")
|
||||
|
|
@ -349,7 +349,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_allclose(output_expected, res[0], rtol=1e-05, atol=1e-08)
|
||||
|
||||
def testRunModel2Contiguous(self):
|
||||
sess = onnxrt.InferenceSession(get_name("matmul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("matmul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array([[2.0, 1.0], [4.0, 3.0], [6.0, 5.0]], dtype=np.float32)[:, [1, 0]]
|
||||
input_name = sess.get_inputs()[0].name
|
||||
self.assertEqual(input_name, "X")
|
||||
|
|
@ -378,7 +378,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
so = onnxrt.SessionOptions()
|
||||
so.log_verbosity_level = 1
|
||||
so.logid = "MultiThreadsTest"
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so)
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so, providers=available_providers)
|
||||
ro1 = onnxrt.RunOptions()
|
||||
ro1.logid = "thread1"
|
||||
t1 = threading.Thread(target=self.run_model, args=(sess, ro1))
|
||||
|
|
@ -391,7 +391,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
t2.join()
|
||||
|
||||
def testListAsInput(self):
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32)
|
||||
input_name = sess.get_inputs()[0].name
|
||||
res = sess.run([], {input_name: x.tolist()})
|
||||
|
|
@ -399,7 +399,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_allclose(output_expected, res[0], rtol=1e-05, atol=1e-08)
|
||||
|
||||
def testStringListAsInput(self):
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array(['this', 'is', 'identity', 'test'], dtype=str).reshape((2, 2))
|
||||
x_name = sess.get_inputs()[0].name
|
||||
res = sess.run([], {x_name: x.tolist()})
|
||||
|
|
@ -410,7 +410,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
self.assertTrue('CPU' in device or 'GPU' in device)
|
||||
|
||||
def testRunModelSymbolicInput(self):
|
||||
sess = onnxrt.InferenceSession(get_name("matmul_2.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("matmul_2.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32)
|
||||
input_name = sess.get_inputs()[0].name
|
||||
self.assertEqual(input_name, "X")
|
||||
|
|
@ -427,7 +427,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_allclose(output_expected, res[0], rtol=1e-05, atol=1e-08)
|
||||
|
||||
def testBooleanInputs(self):
|
||||
sess = onnxrt.InferenceSession(get_name("logicaland.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("logicaland.onnx"), providers=onnxrt.get_available_providers())
|
||||
a = np.array([[True, True], [False, False]], dtype=bool)
|
||||
b = np.array([[True, False], [True, False]], dtype=bool)
|
||||
|
||||
|
|
@ -459,7 +459,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_equal(output_expected, res[0])
|
||||
|
||||
def testStringInput1(self):
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array(['this', 'is', 'identity', 'test'], dtype=str).reshape((2, 2))
|
||||
|
||||
x_name = sess.get_inputs()[0].name
|
||||
|
|
@ -480,7 +480,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_equal(x, res[0])
|
||||
|
||||
def testStringInput2(self):
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array(['Olá', '你好', '여보세요', 'hello'], dtype=str).reshape((2, 2))
|
||||
|
||||
x_name = sess.get_inputs()[0].name
|
||||
|
|
@ -501,7 +501,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_equal(x, res[0])
|
||||
|
||||
def testInputBytes(self):
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array([b'this', b'is', b'identity', b'test']).reshape((2, 2))
|
||||
|
||||
x_name = sess.get_inputs()[0].name
|
||||
|
|
@ -522,7 +522,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_equal(x, res[0].astype('|S8'))
|
||||
|
||||
def testInputObject(self):
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"), providers=onnxrt.get_available_providers())
|
||||
x = np.array(['this', 'is', 'identity', 'test'], object).reshape((2, 2))
|
||||
|
||||
x_name = sess.get_inputs()[0].name
|
||||
|
|
@ -543,7 +543,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_equal(x, res[0])
|
||||
|
||||
def testInputVoid(self):
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("identity_string.onnx"), providers=onnxrt.get_available_providers())
|
||||
# numpy 1.20+ doesn't automatically pad the bytes based entries in the array when dtype is np.void,
|
||||
# so we use inputs where that is the case
|
||||
x = np.array([b'must', b'have', b'same', b'size'], dtype=np.void).reshape((2, 2))
|
||||
|
|
@ -569,7 +569,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
|
||||
def testRaiseWrongNumInputs(self):
|
||||
with self.assertRaises(ValueError) as context:
|
||||
sess = onnxrt.InferenceSession(get_name("logicaland.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("logicaland.onnx"), providers=onnxrt.get_available_providers())
|
||||
a = np.array([[True, True], [False, False]], dtype=bool)
|
||||
res = sess.run([], {'input:0': a})
|
||||
|
||||
|
|
@ -579,7 +579,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
model_path = "../models/opset8/test_squeezenet/model.onnx"
|
||||
if not os.path.exists(model_path):
|
||||
return
|
||||
sess = onnxrt.InferenceSession(model_path)
|
||||
sess = onnxrt.InferenceSession(model_path, providers=onnxrt.get_available_providers())
|
||||
modelmeta = sess.get_modelmeta()
|
||||
self.assertEqual('onnx-caffe2', modelmeta.producer_name)
|
||||
self.assertEqual('squeezenet_old', modelmeta.graph_name)
|
||||
|
|
@ -590,7 +590,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
def testProfilerWithSessionOptions(self):
|
||||
so = onnxrt.SessionOptions()
|
||||
so.enable_profiling = True
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so)
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so,
|
||||
providers=onnxrt.get_available_providers())
|
||||
x = np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32)
|
||||
sess.run([], {'X': x})
|
||||
profile_file = sess.end_profiling()
|
||||
|
|
@ -608,7 +609,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
def getSingleSessionProfilingStartTime():
|
||||
so = onnxrt.SessionOptions()
|
||||
so.enable_profiling = True
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so)
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so,
|
||||
providers=onnxrt.get_available_providers())
|
||||
return sess.get_profiling_start_time_ns()
|
||||
|
||||
# Get 1st profiling's start time
|
||||
|
|
@ -627,14 +629,16 @@ class TestInferenceSession(unittest.TestCase):
|
|||
self.assertEqual(opt.graph_optimization_level, onnxrt.GraphOptimizationLevel.ORT_ENABLE_ALL)
|
||||
opt.graph_optimization_level = onnxrt.GraphOptimizationLevel.ORT_ENABLE_EXTENDED
|
||||
self.assertEqual(opt.graph_optimization_level, onnxrt.GraphOptimizationLevel.ORT_ENABLE_EXTENDED)
|
||||
sess = onnxrt.InferenceSession(get_name("logicaland.onnx"), sess_options=opt)
|
||||
sess = onnxrt.InferenceSession(get_name("logicaland.onnx"), sess_options=opt,
|
||||
providers=onnxrt.get_available_providers())
|
||||
a = np.array([[True, True], [False, False]], dtype=bool)
|
||||
b = np.array([[True, False], [True, False]], dtype=bool)
|
||||
|
||||
res = sess.run([], {'input1:0': a, 'input:0': b})
|
||||
|
||||
def testSequenceLength(self):
|
||||
sess = onnxrt.InferenceSession(get_name("sequence_length.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("sequence_length.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
x = [
|
||||
np.array([1.0, 0.0, 3.0, 44.0, 23.0, 11.0], dtype=np.float32).reshape((2, 3)),
|
||||
np.array([1.0, 0.0, 3.0, 44.0, 23.0, 11.0], dtype=np.float32).reshape((2, 3))
|
||||
|
|
@ -655,7 +659,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
self.assertEqual(output_expected, res[0])
|
||||
|
||||
def testSequenceConstruct(self):
|
||||
sess = onnxrt.InferenceSession(get_name("sequence_construct.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("sequence_construct.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
|
||||
self.assertEqual(sess.get_inputs()[0].type, 'tensor(int64)')
|
||||
self.assertEqual(sess.get_inputs()[1].type, 'tensor(int64)')
|
||||
|
|
@ -684,7 +689,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
def testSequenceInsert(self):
|
||||
opt = onnxrt.SessionOptions()
|
||||
opt.execution_mode = onnxrt.ExecutionMode.ORT_SEQUENTIAL
|
||||
sess = onnxrt.InferenceSession(get_name("sequence_insert.onnx"), sess_options=opt)
|
||||
sess = onnxrt.InferenceSession(get_name("sequence_insert.onnx"), sess_options=opt,
|
||||
providers=onnxrt.get_available_providers())
|
||||
|
||||
self.assertEqual(sess.get_inputs()[0].type, 'seq(tensor(int64))')
|
||||
self.assertEqual(sess.get_inputs()[1].type, 'tensor(int64)')
|
||||
|
|
@ -713,7 +719,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
def testLoadingSessionOptionsFromModel(self):
|
||||
try:
|
||||
os.environ['ORT_LOAD_CONFIG_FROM_MODEL'] = str(1)
|
||||
sess = onnxrt.InferenceSession(get_name("model_with_valid_ort_config_json.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("model_with_valid_ort_config_json.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
session_options = sess.get_session_options()
|
||||
|
||||
self.assertEqual(session_options.inter_op_num_threads, 5) # from the ORT config
|
||||
|
|
@ -739,7 +746,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
so = onnxrt.SessionOptions()
|
||||
so.add_free_dimension_override_by_denotation("DATA_BATCH", 3)
|
||||
so.add_free_dimension_override_by_denotation("DATA_CHANNEL", 5)
|
||||
sess = onnxrt.InferenceSession(get_name("abs_free_dimensions.onnx"), so)
|
||||
sess = onnxrt.InferenceSession(get_name("abs_free_dimensions.onnx"), sess_options=so,
|
||||
providers=onnxrt.get_available_providers())
|
||||
input_name = sess.get_inputs()[0].name
|
||||
self.assertEqual(input_name, "x")
|
||||
input_shape = sess.get_inputs()[0].shape
|
||||
|
|
@ -750,7 +758,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
so = onnxrt.SessionOptions()
|
||||
so.add_free_dimension_override_by_name("Dim1", 4)
|
||||
so.add_free_dimension_override_by_name("Dim2", 6)
|
||||
sess = onnxrt.InferenceSession(get_name("abs_free_dimensions.onnx"), so)
|
||||
sess = onnxrt.InferenceSession(get_name("abs_free_dimensions.onnx"), sess_options=so,
|
||||
providers=onnxrt.get_available_providers())
|
||||
input_name = sess.get_inputs()[0].name
|
||||
self.assertEqual(input_name, "x")
|
||||
input_shape = sess.get_inputs()[0].shape
|
||||
|
|
@ -784,7 +793,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
# Create an InferenceSession that only uses the CPU EP and validate that it uses the
|
||||
# initializer provided via the SessionOptions instance (overriding the model initializer)
|
||||
# We only use the CPU EP because the initializer we created is on CPU and we want the model to use that
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), so, ['CPUExecutionProvider'])
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so, providers=['CPUExecutionProvider'])
|
||||
res = sess.run(["Y"], {"X": np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32)})
|
||||
self.assertTrue(np.array_equal(res[0], np.array([[2.0, 2.0], [12.0, 12.0], [30.0, 30.0]], dtype=np.float32)))
|
||||
|
||||
|
|
@ -812,8 +821,10 @@ class TestInferenceSession(unittest.TestCase):
|
|||
so1 = onnxrt.SessionOptions()
|
||||
so1.register_custom_ops_library(shared_library)
|
||||
|
||||
available_providers = onnxrt.get_available_providers()
|
||||
|
||||
# Model loading successfully indicates that the custom op node could be resolved successfully
|
||||
sess1 = onnxrt.InferenceSession(custom_op_model, so1)
|
||||
sess1 = onnxrt.InferenceSession(custom_op_model, sess_options=so1, providers=available_providers)
|
||||
#Run with input data
|
||||
input_name_0 = sess1.get_inputs()[0].name
|
||||
input_name_1 = sess1.get_inputs()[1].name
|
||||
|
|
@ -829,12 +840,12 @@ class TestInferenceSession(unittest.TestCase):
|
|||
so2 = so1
|
||||
|
||||
# Model loading successfully indicates that the custom op node could be resolved successfully
|
||||
sess2 = onnxrt.InferenceSession(custom_op_model, so2)
|
||||
sess2 = onnxrt.InferenceSession(custom_op_model, sess_options=so2, providers=available_providers)
|
||||
|
||||
# Create another SessionOptions instance with the same shared library referenced
|
||||
so3 = onnxrt.SessionOptions()
|
||||
so3.register_custom_ops_library(shared_library)
|
||||
sess3 = onnxrt.InferenceSession(custom_op_model, so3)
|
||||
sess3 = onnxrt.InferenceSession(custom_op_model, sess_options=so3, providers=available_providers)
|
||||
|
||||
def testOrtValue(self):
|
||||
|
||||
|
|
@ -842,7 +853,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
numpy_arr_output = np.array([[1.0, 4.0], [9.0, 16.0], [25.0, 36.0]], dtype=np.float32)
|
||||
|
||||
def test_session_with_ortvalue_input(ortvalue):
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mul_1.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
res = sess.run(["Y"], {"X": ortvalue})
|
||||
self.assertTrue(np.array_equal(res[0], numpy_arr_output))
|
||||
|
||||
|
|
@ -1009,12 +1021,12 @@ class TestInferenceSession(unittest.TestCase):
|
|||
so1 = onnxrt.SessionOptions()
|
||||
so1.log_severity_level = 1
|
||||
so1.add_session_config_entry("session.use_env_allocators", "1");
|
||||
onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so1)
|
||||
onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so1, providers=onnxrt.get_available_providers())
|
||||
|
||||
# Create a session that will NOT use the registered arena based allocator
|
||||
so2 = onnxrt.SessionOptions()
|
||||
so2.log_severity_level = 1
|
||||
onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so2)
|
||||
onnxrt.InferenceSession(get_name("mul_1.onnx"), sess_options=so2, providers=onnxrt.get_available_providers())
|
||||
|
||||
def testCheckAndNormalizeProviderArgs(self):
|
||||
from onnxruntime.capi.onnxruntime_inference_collection import check_and_normalize_provider_args
|
||||
|
|
|
|||
|
|
@ -39,7 +39,7 @@ class TestBackend(unittest.TestCase):
|
|||
This test is to ensure that the case is covered.
|
||||
"""
|
||||
name = get_name("alloc_tensor_reuse.onnx")
|
||||
sess = onnxrt.InferenceSession(name)
|
||||
sess = onnxrt.InferenceSession(name, providers=onnxrt.get_available_providers())
|
||||
|
||||
run_options = onnxrt.RunOptions()
|
||||
run_options.only_execute_path_to_fetches = True
|
||||
|
|
|
|||
|
|
@ -1,5 +1,5 @@
|
|||
import numpy as np
|
||||
import onnxruntime
|
||||
import onnxruntime as onnxrt
|
||||
import unittest
|
||||
|
||||
from helper import get_name
|
||||
|
|
@ -7,13 +7,13 @@ from helper import get_name
|
|||
class TestIOBinding(unittest.TestCase):
|
||||
|
||||
def create_ortvalue_input_on_gpu(self):
|
||||
return onnxruntime.OrtValue.ortvalue_from_numpy(np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32), 'cuda', 0)
|
||||
return onnxrt.OrtValue.ortvalue_from_numpy(np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32), 'cuda', 0)
|
||||
|
||||
def create_ortvalue_alternate_input_on_gpu(self):
|
||||
return onnxruntime.OrtValue.ortvalue_from_numpy(np.array([[2.0, 4.0], [6.0, 8.0], [10.0, 12.0]], dtype=np.float32), 'cuda', 0)
|
||||
return onnxrt.OrtValue.ortvalue_from_numpy(np.array([[2.0, 4.0], [6.0, 8.0], [10.0, 12.0]], dtype=np.float32), 'cuda', 0)
|
||||
|
||||
def create_uninitialized_ortvalue_input_on_gpu(self):
|
||||
return onnxruntime.OrtValue.ortvalue_from_shape_and_type([3, 2], np.float32, 'cuda', 0)
|
||||
return onnxrt.OrtValue.ortvalue_from_shape_and_type([3, 2], np.float32, 'cuda', 0)
|
||||
|
||||
def create_numpy_input(self):
|
||||
return np.array([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], dtype=np.float32)
|
||||
|
|
@ -27,7 +27,7 @@ class TestIOBinding(unittest.TestCase):
|
|||
def test_bind_input_to_cpu_arr(self):
|
||||
input = self.create_numpy_input()
|
||||
|
||||
session = onnxruntime.InferenceSession(get_name("mul_1.onnx"))
|
||||
session = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
io_binding = session.io_binding()
|
||||
|
||||
# Bind Numpy object (input) that's on CPU to wherever the model needs it
|
||||
|
|
@ -48,7 +48,7 @@ class TestIOBinding(unittest.TestCase):
|
|||
def test_bind_input_only(self):
|
||||
input = self.create_ortvalue_input_on_gpu()
|
||||
|
||||
session = onnxruntime.InferenceSession(get_name("mul_1.onnx"))
|
||||
session = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
io_binding = session.io_binding()
|
||||
|
||||
# Bind input to CUDA
|
||||
|
|
@ -69,7 +69,7 @@ class TestIOBinding(unittest.TestCase):
|
|||
def test_bind_input_and_preallocated_output(self):
|
||||
input = self.create_ortvalue_input_on_gpu()
|
||||
|
||||
session = onnxruntime.InferenceSession(get_name("mul_1.onnx"))
|
||||
session = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
io_binding = session.io_binding()
|
||||
|
||||
# Bind input to CUDA
|
||||
|
|
@ -95,7 +95,7 @@ class TestIOBinding(unittest.TestCase):
|
|||
|
||||
|
||||
def test_bind_input_and_non_preallocated_output(self):
|
||||
session = onnxruntime.InferenceSession(get_name("mul_1.onnx"))
|
||||
session = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
io_binding = session.io_binding()
|
||||
|
||||
# Bind input to CUDA
|
||||
|
|
@ -136,7 +136,7 @@ class TestIOBinding(unittest.TestCase):
|
|||
self.assertTrue(np.array_equal(self.create_expected_output_alternate(), ort_outputs[0].numpy()))
|
||||
|
||||
def test_bind_input_and_bind_output_with_ortvalues(self):
|
||||
session = onnxruntime.InferenceSession(get_name("mul_1.onnx"))
|
||||
session = onnxrt.InferenceSession(get_name("mul_1.onnx"), providers=onnxrt.get_available_providers())
|
||||
io_binding = session.io_binding()
|
||||
|
||||
# Bind ortvalue as input
|
||||
|
|
|
|||
|
|
@ -11,7 +11,8 @@ from helper import get_name
|
|||
class TestInferenceSession(unittest.TestCase):
|
||||
|
||||
def testZipMapStringFloat(self):
|
||||
sess = onnxrt.InferenceSession(get_name("zipmap_stringfloat.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("zipmap_stringfloat.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
x = np.array([1.0, 0.0, 3.0, 44.0, 23.0, 11.0], dtype=np.float32).reshape((2, 3))
|
||||
|
||||
x_name = sess.get_inputs()[0].name
|
||||
|
|
@ -37,7 +38,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
self.assertEqual(output_expected, res[0])
|
||||
|
||||
def testZipMapInt64Float(self):
|
||||
sess = onnxrt.InferenceSession(get_name("zipmap_int64float.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("zipmap_int64float.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
x = np.array([1.0, 0.0, 3.0, 44.0, 23.0, 11.0], dtype=np.float32).reshape((2, 3))
|
||||
|
||||
x_name = sess.get_inputs()[0].name
|
||||
|
|
@ -55,7 +57,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
self.assertEqual(output_expected, res[0])
|
||||
|
||||
def testDictVectorizer(self):
|
||||
sess = onnxrt.InferenceSession(get_name("pipeline_vectorize.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("pipeline_vectorize.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
input_name = sess.get_inputs()[0].name
|
||||
self.assertEqual(input_name, "float_input")
|
||||
input_type = str(sess.get_inputs()[0].type)
|
||||
|
|
@ -99,7 +102,8 @@ class TestInferenceSession(unittest.TestCase):
|
|||
np.testing.assert_allclose(output_expected, res[0], rtol=1e-05, atol=1e-08)
|
||||
|
||||
def testLabelEncoder(self):
|
||||
sess = onnxrt.InferenceSession(get_name("LabelEncoder.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("LabelEncoder.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
input_name = sess.get_inputs()[0].name
|
||||
self.assertEqual(input_name, "input")
|
||||
input_type = str(sess.get_inputs()[0].type)
|
||||
|
|
@ -141,7 +145,7 @@ class TestInferenceSession(unittest.TestCase):
|
|||
sess = onnxrt.InferenceSession(get_name("mlnet_encoder.onnx"), None,
|
||||
['DmlExecutionProvider', 'CPUExecutionProvider'])
|
||||
else:
|
||||
sess = onnxrt.InferenceSession(get_name("mlnet_encoder.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("mlnet_encoder.onnx"), providers=available_providers)
|
||||
|
||||
names = [_.name for _ in sess.get_outputs()]
|
||||
self.assertEqual(['C00', 'C12'], names)
|
||||
|
|
|
|||
|
|
@ -502,7 +502,7 @@ class TestNuphar(unittest.TestCase):
|
|||
data_seq_len = np.random.randint(1, seq_len, size=(batch_size,), dtype=np.int32)
|
||||
|
||||
# run lstm as baseline
|
||||
sess = onnxrt.InferenceSession(lstm_model_name)
|
||||
sess = onnxrt.InferenceSession(lstm_model_name, providers=onnxrt.get_available_providers())
|
||||
first_lstm_data_output = sess.run([], {'input': data_input[:, 0:1, :], 'seq_len': data_seq_len[0:1]})
|
||||
|
||||
lstm_data_output = []
|
||||
|
|
@ -524,7 +524,7 @@ class TestNuphar(unittest.TestCase):
|
|||
check=True)
|
||||
|
||||
# run scan_batch with batch size 1
|
||||
sess = onnxrt.InferenceSession(scan_model_name)
|
||||
sess = onnxrt.InferenceSession(scan_model_name, providers=onnxrt.get_available_providers())
|
||||
scan_batch_data_output = sess.run([], {'input': data_input[:, 0:1, :], 'seq_len': data_seq_len[0:1]})
|
||||
assert np.allclose(first_lstm_data_output, scan_batch_data_output)
|
||||
|
||||
|
|
@ -579,7 +579,7 @@ class TestNuphar(unittest.TestCase):
|
|||
'--output', matmul_model_name, '--mode', 'gemm_to_matmul'
|
||||
], check=True)
|
||||
|
||||
sess = onnxrt.InferenceSession(matmul_model_name)
|
||||
sess = onnxrt.InferenceSession(matmul_model_name, providers=onnxrt.get_available_providers())
|
||||
test_inputs = {}
|
||||
set_gemm_model_inputs(running_config, test_inputs, a, b, c)
|
||||
actual_y = sess.run([], test_inputs)
|
||||
|
|
@ -651,7 +651,7 @@ class TestNuphar(unittest.TestCase):
|
|||
b1 = b1.reshape((1, ) + b1.shape)
|
||||
a2 = a2.reshape((1, ) + a2.shape)
|
||||
b2 = b2.reshape((1, ) + b2.shape)
|
||||
sess = onnxrt.InferenceSession(gemm_model_name)
|
||||
sess = onnxrt.InferenceSession(gemm_model_name, providers=onnxrt.get_available_providers())
|
||||
|
||||
test_inputs = {}
|
||||
set_gemm_model_inputs(running_config1, test_inputs, a1, b1, c1)
|
||||
|
|
@ -667,7 +667,7 @@ class TestNuphar(unittest.TestCase):
|
|||
], check=True)
|
||||
|
||||
# run after model editing
|
||||
sess = onnxrt.InferenceSession(matmul_model_name)
|
||||
sess = onnxrt.InferenceSession(matmul_model_name, providers=onnxrt.get_available_providers())
|
||||
actual_y = sess.run([], test_inputs)
|
||||
|
||||
assert np.allclose(expected_y, actual_y, atol=1e-7)
|
||||
|
|
|
|||
|
|
@ -24,7 +24,8 @@ class TestSparseToDenseMatmul(unittest.TestCase):
|
|||
dense_shape = [3,3]
|
||||
values = np.array([1.764052391052246, 0.40015721321105957, 0.978738009929657], np.float)
|
||||
indices = np.array([2, 3, 5], np.int64)
|
||||
sess = onnxrt.InferenceSession(get_name("sparse_initializer_as_output.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("sparse_initializer_as_output.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
res = sess.run_with_ort_values(["values"], {})
|
||||
self.assertEqual(len(res), 1)
|
||||
ort_value = res[0]
|
||||
|
|
@ -83,7 +84,8 @@ class TestSparseToDenseMatmul(unittest.TestCase):
|
|||
786, 915, 1044, 3324, 3462, 3600, 4911, 5178, 5445,
|
||||
894, 1041, 1188, 3756, 3912, 4068, 5559, 5862, 6165], np.float).reshape(common_shape)
|
||||
|
||||
sess = onnxrt.InferenceSession(get_name("sparse_to_dense_matmul.onnx"))
|
||||
sess = onnxrt.InferenceSession(get_name("sparse_to_dense_matmul.onnx"),
|
||||
providers=onnxrt.get_available_providers())
|
||||
res = sess.run_with_ort_values(["dense_Y"], { "sparse_A" : A_ort_value, "dense_B" : B_ort_value })
|
||||
self.assertEqual(len(res), 1)
|
||||
ort_value = res[0]
|
||||
|
|
|
|||
|
|
@ -80,7 +80,8 @@ class ONNXExporterTest(unittest.TestCase):
|
|||
custom_opsets=custom_opsets)
|
||||
|
||||
# compute onnxruntime output prediction
|
||||
ort_sess = onnxruntime.InferenceSession(f.getvalue())
|
||||
ort_sess = onnxruntime.InferenceSession(f.getvalue(),
|
||||
providers=onnxruntime.get_available_providers())
|
||||
input_copy = copy.deepcopy(input)
|
||||
ort_test_with_input(ort_sess, input_copy, output, rtol, atol)
|
||||
|
||||
|
|
|
|||
|
|
@ -280,7 +280,8 @@ def generate_test_data(onnx_file,
|
|||
else:
|
||||
sess_options.optimized_model_filepath = path_prefix + "_optimized_gpu.onnx"
|
||||
|
||||
session = onnxruntime.InferenceSession(onnx_file, sess_options)
|
||||
session = onnxruntime.InferenceSession(onnx_file, sess_options=sess_options,
|
||||
providers=onnxruntime.get_available_providers())
|
||||
if use_cpu:
|
||||
session.set_providers(['CPUExecutionProvider']) # use cpu
|
||||
else:
|
||||
|
|
|
|||
|
|
@ -20,6 +20,11 @@ class TrainingSession(InferenceSession):
|
|||
else:
|
||||
self._sess = C.TrainingSession()
|
||||
|
||||
# providers needs to be passed explicitly as of ORT 1.10
|
||||
# retain the pre-1.10 behavior by setting to the available providers.
|
||||
if providers is None:
|
||||
providers = C.get_available_providers()
|
||||
|
||||
providers, provider_options = check_and_normalize_provider_args(providers, provider_options,
|
||||
C.get_available_providers())
|
||||
|
||||
|
|
|
|||
|
|
@ -315,7 +315,7 @@ class ORTTrainer(object):
|
|||
training_mode_node.attribute[0].name = "value"
|
||||
ratio_node.attribute[0].name = "value"
|
||||
|
||||
_inference_sess = ort.InferenceSession(onnx_model_copy.SerializeToString())
|
||||
_inference_sess = ort.InferenceSession(onnx_model_copy.SerializeToString(), providers=ort.get_available_providers())
|
||||
inf_inputs = {}
|
||||
for i, input_elem in enumerate(input):
|
||||
inf_inputs[_inference_sess.get_inputs()[i].name] = input_elem.cpu().numpy()
|
||||
|
|
|
|||
Loading…
Reference in a new issue