microsoft / microsoft/onnxruntime
[Feature Request] [WebGPU EP] Enable more graph fusions
- Dominant language
- C++
- Stars
- 21.9k
- Forks
- 4.2k
- Avg merge
- 4d 11h
- Merged PRs (30d)
- 184
Description
### Describe the feature request
Investigate and enable graph fusions already supported by WebGPU EP.
Some graph fusions to check:
* Gelu
* FastGelu
* BiasGelu
* LayerNorm
* SimplifiedLayerNorm
* SkipLayerNorm
**Possible changes needed**:
* In `onnxruntime/core/optimizer/graph_transformer_utils.cc`, in `GenerateTransformers()`, enable the fusions for WebGPU EP.
* Make sure the changes are well covered by tests.
* Something else?
### Describe scenario use case
Any model whose inference time can benefit from the fusions
Contributor guide
Research direction
Start in onnxruntime/core/optimizer/graph_transformer_utils.cc, specifically GenerateTransformers(), and inspect how WebGPU EP enables graph fusions. Check Gelu, FastGelu, BiasGelu, LayerNorm, SimplifiedLayerNorm, and SkipLayerNorm, then identify the existing test coverage for these transformations. Done means the supported fusions are enabled for WebGPU EP and are covered by tests.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- machine-learning, performance
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 55/100