Lunderberg opened a new pull request, #15562: URL: https://github.com/apache/tvm/pull/15562
Prior to this commit, the `relax.concat` operator determined the concatenated dimension by adding together the dimension across all concatenated values. For shapes involving symbolic variables, this can produce less readable shapes in some cases. For example, concatenating together four arrays, each of shape `[128 // factor]` would produce an output of `[128 // factor + 128 // factor + 128 // factor + 128 // factor]`. This commit updates the shape inference of `relax.concat` to check for the common case of equal-sized tensors being concatenated. In the example above, the shape would instead be inferred as `[(128 // factor) * 4]`. -- 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]
