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
As a MIGraphX performance developer I want to break down models in to fused portions and run the single fusion in isolation using different graph optimizers.
Now that the MIGX IR can be extracted in to files PR (#3550) , the next step is to convert the IR back to ONNX. This is very important as it allows for the performance comparison with other ONNX accepting optimizers. Comparing MIGX on a per ONNX operator basis is not as interesting as comparing on a per fusion basis
The ONNX community should have tools to build an ONNX model that will likely come in handy.
It may become easier to create a MIGX api to generate the ONNX Operators but feel free to use the newly generated mlir solutions as your starting point.
Success
Convert each .mlir file to ONNX and run perf results for both migraphx-driver and trtexec
The text was updated successfully, but these errors were encountered:
Desciption
As a MIGraphX performance developer I want to break down models in to fused portions and run the single fusion in isolation using different graph optimizers.
Now that the MIGX IR can be extracted in to files PR (#3550) , the next step is to convert the IR back to ONNX. This is very important as it allows for the performance comparison with other ONNX accepting optimizers. Comparing MIGX on a per ONNX operator basis is not as interesting as comparing on a per fusion basis
The ONNX community should have tools to build an ONNX model that will likely come in handy.
It may become easier to create a MIGX api to generate the ONNX Operators but feel free to use the newly generated mlir solutions as your starting point.
Success
Convert each .mlir file to ONNX and run perf results for both migraphx-driver and trtexec
The text was updated successfully, but these errors were encountered: