chengven027-intellif commented on code in PR #13747:
URL: https://github.com/apache/tvm/pull/13747#discussion_r1066501716
##########
python/tvm/relay/frontend/onnx.py:
##########
@@ -4898,6 +4898,90 @@ def _impl_v10(cls, inputs, attr, params):
return out
+class QGemm(OnnxOpConverter):
+ """Operator converter for QGemm."""
+
+ @classmethod
+ def _impl_v1(cls, inputs, attr, params):
+ #
http://www.xavierdupre.fr/app/mlprodict/helpsphinx/onnxops/onnx_commicrosoft_QGemm.html
+
+ a = inputs[0]
+ a_scale = get_scalar(inputs[1], params)
+ a_zp = get_scalar(inputs[2], params, "int32")
+
+ b = inputs[3]
+ # a scalar or 1D tensor of size 1 or N
+ b_scale = get_scalar_or_1d_tensor(inputs[4], params)
+ # a scalar or 1D tensor of size 1 or N
Review Comment:
yes，tensors `A scale and zero point` do not have `N`，Itis a scalar. And
tensor `B scale and zero point` can have N size 1-D tensor, which means
per-column quantization.
--
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]
