Closed neCo2 closed 5 months ago
parseEventStream
's only job is to return the event stream as json, and it doesn't look like any backend sends back token ids in the completion stream. So I feel like the only choice would be to detokenize each token.
I see, that's a shame. Guess I'll have to figure something out for kobold then.
Started working on phrase biasing, and being able to monitor the stream of token ids instead of the token strings would make life much easier for me. I'd do it myself, but I couldn't figure it out.