cbalint13 commented on PR #19863: URL: https://github.com/apache/tvm/pull/19863#issuecomment-4767445110
@guan404ming , @tlopex Cc @Hzfengsy , @msr-h, @tqchen Perhaps not here is the best discussion for this (out of context), but I rise it here: Why testcases did not use onnx own provided [ReferenceEvaluator](https://onnx.ai/onnx/api/reference.html) ? To my understanding (and some experience) ```ReferenceEvaluator``` is not only for functional correctness (indeed onnxruntime also does it) but provides specification compliance too. It comes with onnx package itself, provides coverage for operator histories too. Also any particular operator can be also sweeped over its all existing op versions using history schema, (see a random testcase [example](https://github.com/cbalint13/onnx2mlir/blob/6df5c9d36fa5de6acf883fdc10df5bcc910e035f/tests/python/conversion/test_onnx_to_linalg_ops.py#L289-L345)) . ONNX really excels at reflecting its own schema with reference validation too. I am not saying we should drop onnxruntime now, but if we want excelent onnx coverage with full history schema and compliance we may consider this. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
