Zheng-Bicheng commented on PR #15417: URL: https://github.com/apache/tvm/pull/15417#issuecomment-1663322791
@jiangjiajun The `quantize_linear` operator of `paddle` seems to encounter calculation errors after being transformed into `tvm's` `_qnn.op.dequantize`, particularly in extreme cases. I am unsure if this issue is related to TVM optimization. Is there a way to investigate this? I tested the model used in CI with data ranging from -1 to 1, and the average cosine similarity is around **98%**. -- 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]
