Prepare for new models in roadmap: rnnt

[mrasquinha-g]

As discussed in the meeting on 1/24, we envision having to retire benchmarks in upcoming submissions. This discussion is to identify potential candidates. Please share your feedback on rnnt being a potential candidate.

[psyhtest]

While Speech Recognition is useful in general, the current reference implementation is not quantization friendly, in our (admittedly limited) experience.

[arjunsuresh]

@mrasquinha-g Instead of having different discussions for individual models can we have a single discussion for all the models in it?