microsoft / microsoft/onnxruntime

Onnx model failing on main branch. But working with rel-1.14.1 branch

Open
#16,188 4 comments 3 reactions 1 assignee Claimed by @yufenglee View on GitHub
quantization
Dominant language
C++
Stars
21.9k
Forks
4.2k
Avg merge
4d 11h
Merged PRs (30d)
184

Description

### Describe the issue

The attached onnx file could not be inferenced successfully by ORT 1.15 (main branch) while the previous ORT (rel1.14.1 branch) works well.

Quantized Onnx model: [osnet_ain_x1_0_vehicle_reid.zip](https://github.com/microsoft/onnxruntime/files/11621976/osnet_ain_x1_0_vehicle_reid.zip)

### To reproduce

git clone https://github.com/microsoft/onnxruntime.git
cd onnxruntime
git checkout rel-1.14.1 ===>the onnx model infers successfully

git checkout main ===> the onnx model fails with the following error:

_Run failed:PerformanceRunner::RunOneIteration caught exception: Non-zero status code returned while running DequantizeLinear node. Name:'DequantizeLinear_conv2.0.gate.fc2.weight_1/duplicated' Status Message: /home/icx-8380/saurabh/vehicle_reid/onnxruntime/onnxruntime/core/providers/cpu/quantization/quantize_linear.cc:37 void onnxruntime::PrepareForQDQ(const onnxruntime::TensorShape&, const onnxruntime::Tensor&, const onnxruntime::Tensor*, int64_t, int64_t&, int64_t&, int64_t&) scale.Shape().NumDimensions() == 1 && scale.Shape()[0] == broadcast_dim was false. scale must be 1D tensor with size 4_

Used ./onnxruntime_perf_test for reproducing

Command used: ./onnxruntime_perf_test -m times -r 1

onnx model file:
[osnet_ain_x1_0_vehicle_reid.zip](https://github.com/microsoft/onnxruntime/files/11621955/osnet_ain_x1_0_vehicle_reid.zip)

### Urgency

_No response_

### Platform

Linux

### OS Version

Ubuntu 20.04.5 LTS

### ONNX Runtime Installation

Built from Source

### ONNX Runtime Version or Commit ID

issue on main branch on 1st June 2023

### ONNX Runtime API

C++

### Architecture

X64

### Execution Provider

Default CPU

### Execution Provider Library Version

_No response_

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.