Closed Subuday closed 9 months ago
Hello!
The conformer there was only experimental and was not used in the experiments or the paper. I only left it there to give an example in case somebody wants to switch it with something else (ex. Conformer, Mamba, RWKV etc. or whatever comes in the future)
I think you can modify how the timestep is added to the hidden states manually if you want to fix it.
Regards, Shivam
Thanks.
Why when Conformer is used, time embeddings are not incorporated?