microsoft / microsoft/onnxruntime
[Feature Request] Consider support int4/uint4 for reshape op of default CPU EP
Open
feature request
WebNN
- Dominant language
- C++
- Stars
- 21.9k
- Forks
- 4.2k
- Avg merge
- 4d 11h
- Merged PRs (30d)
- 184
Description
### Describe the feature request
ONNX opset 21 added support for int4 and uint4 of reshape op. So I suppose default CPU EP should support int4 and uint4 data type for reshape op.
### Describe scenario use case
For WebNN use case, we need to insert `reshape` before scale and zeroPoint for dequantizeLinear implementation since ONNX constrains scale and zeroPoint(if exits) must be 1D tensor for per-axis dequantizeLinear. In some scenarios, the data type of zero_point is int4 or uint4.
Contributor guide
Assessment
This issue has not been assessed yet.