handar423 commented on a change in pull request #5695:
URL: https://github.com/apache/incubator-tvm/pull/5695#discussion_r432826485
##########
File path: python/tvm/relay/op/_tensor_grad.py
##########
@@ -472,8 +472,8 @@ def bias_add_grad(orig, grad):
def dense_grad(orig, grad):
"""Returns [grad' @ weight, data @ grad']"""
data, weight = orig.args
- return [collapse_sum_like(transpose(grad) * weight, data),
- collapse_sum_like(data * transpose(grad), weight)]
+ return [collapse_sum_like(_nn.dense(grad, transpose(weight)), data),
Review comment:
Thank you for your response!
After correcting it, I found that both x86 and CUDA only support dense
without batching, I tried testing with arm-cpu(mobile) but came into
[python/tvm/relay/op/strategy/x86.py](https://github.com/apache/incubator-tvm/blob/master/python/tvm/relay/op/strategy/x86.py)
and failed as well, so could you please tell me how to run dense with
batching?
Thanks again!
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
