mbrookhart opened a new pull request #7927: URL: https://github.com/apache/tvm/pull/7927
This PR refactors the relay scatter_nd op in two ways: 1) scatter_nd now takes in a data tensor, and updates are scattered based on indices. The previous API assumed zero-initialization for the data. 2) scatter_nd now has a "mode" argument to determine how to update the input data, either "add" or "update". This is to support the use cases as a gradient of gather_nd and to support the ONNX and pytorch APIs. TF also supports "max", "min", and "sub", which are not yet supported here. This also adds ScatterND to the ONNX importer. cc @tkonolige @altanh @jwfromm @masahi -- 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: [email protected]
