-
Notifications
You must be signed in to change notification settings - Fork 3.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
fix to skip node not in graph. #5238
Conversation
fix to skip node not in graph because some network cannot be hybridized with some var unused.
Could you add some test case for it? |
@icemelon9 what kind of tests, please? Failure case I've met with? |
I am going to merge this for now, given it is a change to the nnvm code. However, based on my current impression, I am not really sure if such relaxation should be done, given that node2index is supposed to only contain nodes in the graph, @chinakook would be great if you can comment a bit. |
@tqchen As I'v just tested, It's a op named |
i see, perhaps you also want to confirm with the mxnet community as well to make sure it is the intended behavior cc @szha |
I'm sure some network will be fail to hybridize after this apache/mxnet#15167 PR. So we can add |
fix to skip node not in graph because some network cannot be hybridized with some var unused.
fix to skip node not in graph because some network cannot be hybridized with some var unused.
fix to skip node not in graph because some network cannot be hybridized with some var unused.
This is not right and should be reverted IMO - the problem in the original issue (apache/mxnet#18624) was a genuine bug in the fusion graph pass that generated a cycle in the graph. Omitting such node from indexed graph would result in incorrect result being produced. |
@ptrendx feel free to send a PR to revert the change then |
fix to skip node not in graph because some network cannot be hybridized with some var unused.
Thanks for contributing to TVM! Please refer to guideline https://tvm.apache.org/docs/contribute/ for useful information and tips. After the pull request is submitted, please request code reviews from Reviewers by @ them in the pull request thread.