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
We hit this perf issue when doing Fuzz testing using FileFuzzer tool for Speech Runtime, check below for what we got. Normally it is expected that scenario 1 is faster than scenario 2, but it is too slow for CreateSession() function which is blocking us to do the fuzz testing.
Scenario 1: E2E decoding scenario using Command Prompt Windows => Took about 4 secs in total, CreateSession() for big encoder.quantqint8.onnx file (87.8MB) took less than 2 secs, total time for 4 .onnx files is around 2 secs. Scenario 2: E2E decoding scenario through FileFuzzer tool => Took about 67 secs in total, CreateSession() for big encoder.quantqint8.onnx file (87.8MB) took about 51 secs, total time for 4 .onnx files is about 56 secs.
Notes: we verified with the latest version [v1.20.1], this issue also repro.
To reproduce
Here is a local setup to repro the issue using en-US model:
\MixedRealityFS\PRIVATE\Speech\speechshare\USR\yihche\4ONNXRuntimeComponent\ReproFuzzBug11045
Copy ReproFuzzBug11045 to local machine e.g. at E:\FuzzTesting
Install FileFuzzer 3.0 with MSI files at Software\FileFuzzer 3.0\amd64
2.1 Install FCL.msi first
2.2 Then install FileFuzzer.msi
Default installation path is at C:\Program Files\Microsoft\FileFuzzer
Modify FileFuzzer_ConfigFile.xml for , and to matching your local setup path.
Here are the examples:
E:\FuzzTesting\ReproFuzzBug11045\audio\en-US
E:\FuzzTesting\ReproFuzzBug11045\output-Fuzzer
E:\FuzzTesting\ReproFuzzBug11045\ReproBug11045_SlowIssue.bat
Open FileFuzzer UI as administrator from C:\Program Files\Microsoft\FileFuzzer\FileFuzzerConfig.exe
4.1 Click "Load Settings" and select FileFuzzer_ConfigFile.xml
4.2 Click "Start" button.
Check testing results in path
Urgency
It is a blocking issue for us to do Fuzz testing, and we have to reduce seed and iteration numbers as temporary workaround.
Platform
Windows
OS Version
Windows 11
ONNX Runtime Installation
Released Package
ONNX Runtime Version or Commit ID
v1.15.1
ONNX Runtime API
C++
Architecture
X64
Execution Provider
Default CPU
Execution Provider Library Version
No response
Model File
No response
Is this a quantized model?
Yes
The text was updated successfully, but these errors were encountered:
Describe the issue
We hit this perf issue when doing Fuzz testing using FileFuzzer tool for Speech Runtime, check below for what we got. Normally it is expected that scenario 1 is faster than scenario 2, but it is too slow for CreateSession() function which is blocking us to do the fuzz testing.
Scenario 1: E2E decoding scenario using Command Prompt Windows => Took about 4 secs in total, CreateSession() for big encoder.quantqint8.onnx file (87.8MB) took less than 2 secs, total time for 4 .onnx files is around 2 secs.
Scenario 2: E2E decoding scenario through FileFuzzer tool => Took about 67 secs in total, CreateSession() for big encoder.quantqint8.onnx file (87.8MB) took about 51 secs, total time for 4 .onnx files is about 56 secs.
Notes: we verified with the latest version [v1.20.1], this issue also repro.
To reproduce
Here is a local setup to repro the issue using en-US model:
\MixedRealityFS\PRIVATE\Speech\speechshare\USR\yihche\4ONNXRuntimeComponent\ReproFuzzBug11045
2.1 Install FCL.msi first
2.2 Then install FileFuzzer.msi
Default installation path is at C:\Program Files\Microsoft\FileFuzzer
Here are the examples:
E:\FuzzTesting\ReproFuzzBug11045\audio\en-US
E:\FuzzTesting\ReproFuzzBug11045\output-Fuzzer
E:\FuzzTesting\ReproFuzzBug11045\ReproBug11045_SlowIssue.bat
4.1 Click "Load Settings" and select FileFuzzer_ConfigFile.xml
4.2 Click "Start" button.
Urgency
It is a blocking issue for us to do Fuzz testing, and we have to reduce seed and iteration numbers as temporary workaround.
Platform
Windows
OS Version
Windows 11
ONNX Runtime Installation
Released Package
ONNX Runtime Version or Commit ID
v1.15.1
ONNX Runtime API
C++
Architecture
X64
Execution Provider
Default CPU
Execution Provider Library Version
No response
Model File
No response
Is this a quantized model?
Yes
The text was updated successfully, but these errors were encountered: