Mimi Encoder

It seems like Mimi doesn't work on Cuda and using it via cpu requires 1 second to transcribe 80 ms of audio.
Why is this the case? Thanks.
It did mention something along the lines for Cuda there is an missing cuda implementation for rot emb i not implemented.

Start by locating the Mimi encoder and its rotary-embedding path in the Candle repository, then reproduce the reported CUDA failure and CPU timing. Compare the CUDA and CPU execution paths and verify that the encoder can process the reported 80 ms audio segment at a reasonable speed without the missing implementation error.

Written by the indexing model from the issue text.