You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
If the index is out-of-bound, CPU EP will throw runtime error like: Non-zero status code returned while running Gather node. Name:'gather_0' Status Message: indices element out of data bounds, idx=3 must be within the inclusive range [-3,2].
OpenVINO GPU EP won't throw error but get zeros for these indices.
Edited:
Strangely, I tested this case again and the result changed. Not sure if it's caused by the environment change or I made a mistake🤔
OV GPU would fail with runtime error
[28140:15376:0218/141513.353:ERROR:graph_impl_ort.cc(308)] : [WebNN] Failed to call GetOrtModelBuilderApi()->CreateSessionFromModel( env, model_info->model, session_options, session.GetAddressOf()): Exception during initialization: D:\onnxruntime\onnxruntime\core\providers\openvino\backend_manager.cc:163 __cdecl onnxruntime::openvino_ep::BackendManager::BackendManager(const struct onnxruntime::openvino_ep::GlobalContext &,const struct onnxruntime::Node &,const class onnxruntime::GraphViewer &,const struct onnxruntime::logging::Logger &,class onnxruntime::openvino_ep::EPCtxHandler &) D:\onnxruntime\onnxruntime\core\providers\openvino\ov_interface.cc:106 class onnxruntime::openvino_ep::OVExeNetwork __cdecl onnxruntime::openvino_ep::OVCore::CompileModel(const class std::basic_string<char,struct std::char_traits,class std::allocator > &,class std::basic_string<char,struct std::char_traits,class std::allocator > &,class std::map<class std::basic_string<char,struct std::char_traits,class std::allocator >,class ov::Any,struct std::less<class std::basic_string<char,struct std::char_traits,class std::allocator > >,class std::allocator<struct std::pair<class std::basic_string<char,struct std::char_traits,class std::allocator > const ,class ov::Any> > > &,const class std::basic_string<char,struct std::char_traits,class std::allocator > &) [OpenVINO-EP] Exception while Loading Network for graph: OpenVINOExecutionProvider_OpenVINO-EP-subgraph_21_0Exception from src/inference/src/cpp/core.cpp:142:
Exception from src/inference/src/dev/plugin.cpp:53:
src/plugins/intel_gpu/src/graph/crop.cpp at line: 253
Error has occured for: gather:Gather_79724_reshape_output/crop_10
input sizes - offsets sizes: [b:-8, f:12, x:1, y:1, z:0, w:0, u:0, v:0, g:0]
reference input sizes sizes: [b:1, f:12, x:1, y:1, z:1, w:1, u:1, v:1, g:1]
All input sizes - offsets dimensions should not be less than reference input sizes dimensions.
Mismatching dimensions: Batch
Invalid Batch offset: exceeds data for output!
And OV CPU would fail due to wrong results
Fail | gather float32 2D tensor and int32 0D out-of-bound positive indices default options | assert_true: assert_array_approx_equals_ulp: test gather float32 actual 0 should be close enough to expected 90.2870101928711 by the acceptable 0 ULP distance, but they have 1119130355 ULP distance expected true got false
Fail | gather float32 2D tensor and int32 0D out-of-bound negative indices default options | assert_true: assert_array_approx_equals_ulp: test gather float32 actual 0 should be close enough to expected -66.05901336669922 by the acceptable 0 ULP distance, but they have 1031528905 ULP distance expected true got false
OV version: 2024.6.0.17404, Debug version
The text was updated successfully, but these errors were encountered:
If the index is out-of-bound, CPU EP will throw runtime error like: Non-zero status code returned while running Gather node. Name:'gather_0' Status Message: indices element out of data bounds, idx=3 must be within the inclusive range [-3,2].
OpenVINO GPU EP won't throw error but get zeros for these indices.Edited:
Strangely, I tested this case again and the result changed. Not sure if it's caused by the environment change or I made a mistake🤔
OV GPU would fail with runtime error
And OV CPU would fail due to wrong results
OV version: 2024.6.0.17404, Debug version
The text was updated successfully, but these errors were encountered: