- Missing ORT WebGPU operators - Source-pinned operator gaps, importance, partial-support restrictions and Chrome WebNN implications. See the detailed report.
- Web and optimized native benchmarks - Run the same ONNX models with static GPU I/O, captured decode and GPU token selection; validate outputs and compare CPU-visible performance.
- Phi-4 performance trend - Rebuild and benchmark historical ORT WebGPU milestones, verify archived data, and maintain the generated report.
Folders and files
| Name | Name | Last commit date | ||
|---|---|---|---|---|