Speculative decoding example in e2e_grpc_speculative_decoding_client.py

[jxchenus]

I am wondering where the sampling logic is to decide whether a draft token is accepted or rejected. Any pointers?

[jxchenus]

Similarly, in the bls example, there are several functions with NotImplementedError(). So I'm confused on the optimized nature of this example.