masahi opened a new pull request, #14841: URL: https://github.com/apache/tvm/pull/14841
diffusers started to use `scaled_dot_product_attention` as of v0.16 in SD VAE. The PT attention op requires a different input format than our attention op, so we need to transpose them. Luckily those transpose can be cancelled since diffusers does transpose on q/k/v before calling `scaled_dot_product_attention` anyway. -- 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]
