Skip to content

C API Locking prevents custom gradient definition #48767

Closed
#48944
Closed
C API Locking prevents custom gradient definition#48767
#48944
Labels
comp:coreissues related to core part of tensorflowstat:contribution welcomeStatus - Contributions welcometype:bugBug

Description

@rnett
rnett
opened on Apr 27, 2021

I am trying to add custom gradient support to the Java bindings (tensorflow/java#292). This is currently being prevented by the fact that TF_AddGradientsWithPrefix requires a lock on the graph, but so do any operation creation methods (i.e. TF_NewOperation), so it is impossible to create ops in gradient functions.

Is there a way I can get around this? I am considering calling TF_NewOperationLocked directly but I assume the locks are there for a reason.

Activity

  1. rnett
    added
    type:othersissues not falling in bug, perfromance, support, build and install or feature
    on Apr 27, 2021
  2. google-ml-butler
    assigned
    saikumarchalla
    on Apr 27, 2021
  3. saikumarchalla
    assigned
    amahendrakar
    and unassigned
    saikumarchalla
    on Apr 27, 2021
  4. amahendrakar
    added
    comp:apisHighlevel API related issues
    type:supportSupport issues
    and removed
    type:othersissues not falling in bug, perfromance, support, build and install or feature
    on Apr 28, 2021
  5. amahendrakar
    assigned
    jvishnuvardhan
    and unassigned
    amahendrakar
    on Apr 28, 2021
  6. jvishnuvardhan
    added
    comp:coreissues related to core part of tensorflow
    type:bugBug
    stat:awaiting tensorflowerStatus - Awaiting response from tensorflower
    and removed
    comp:apisHighlevel API related issues
    type:supportSupport issues
    on Apr 28, 2021
  7. kkimdev
    added
    stat:contribution welcomeStatus - Contributions welcome
    and removed
    stat:awaiting tensorflowerStatus - Awaiting response from tensorflower
    on May 6, 2021

  8. kkimdev commented on May 6, 2021

    @kkimdev
    kkimdev
    on May 6, 2021
    Contributor

    Thanks for reporting the issue!

  9. saxenasaurabh commented on May 6, 2021

    @saxenasaurabh
    saxenasaurabh
    on May 6, 2021
    Member

    Please feel free to send a PR!

  10. rnett commented on May 6, 2021

    @rnett
    rnett
    on May 6, 2021
    ContributorAuthor

    I can, but what's the preferred solution? Providing and exposing a TF_AddGradientsWithPrefixLocked that doesn't lock, or exposing TF_NewOperationLocked and TF_FinishOperationLocked? (See #48815)

  11. saxenasaurabh commented on May 6, 2021

    @saxenasaurabh
    saxenasaurabh
    on May 6, 2021
    Member

    Exposing TF_NewOperationLocked and TF_FinishOperationLocked sgtm.

  12. jvishnuvardhan
    removed their assignment
    on Jun 18, 2021
  13. copybara-service
    closed this as completedin #48944on Jul 15, 2021

  14. google-ml-butler commented on Jul 15, 2021

    @google-ml-butler
    google-ml-butlerbot
    on Jul 15, 2021

    Are you satisfied with the resolution of your issue?
    Yes
    No

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    comp:coreissues related to core part of tensorflowstat:contribution welcomeStatus - Contributions welcometype:bugBug

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions