tiandiao123 edited a comment on pull request #8808: URL: https://github.com/apache/tvm/pull/8808#issuecomment-914543357
> And do we have a unit test for it? > You can try to load a pre-trained model and process inference with the model before & after TRT. Then compare the cosine similarity of the two result, for most model, even though the result may be different in `np.testing.assert_allclose`, the cosine similarity can still be greater than 0.99(should not less than 0.95, or we can assume the TRT result is totally wrong). Hi @jcf94 I have added a test file: https://github.com/tiandiao123/tvm/blob/pr_trt_int8/tests/python/contrib/test_tensorrt_int8_exp.py -- 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]
