akaashrp opened a new pull request, #20128: URL: https://github.com/apache/tvm/pull/20128
PyTorch exported programs may describe input dimensions using expressions derived from other symbolic dimensions, such as `2 * n`. The importer previously recognized these expressions but replaced them with independent TIR variables when constructing the Relax function signature. This discarded the relationship between the dimensions. This PR: - Preserves derived TIR expressions when constructing input tensor shapes - Records range constraints only for direct symbolic variables - Accepts input shape and dtype metadata from either `tensor_meta` or placeholder value metadata - Reports an explicit error when input shape or dtype metadata is unavailable -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
