-
Notifications
You must be signed in to change notification settings - Fork 15.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Generalized streaming control stop_sequences #14782
Generalized streaming control stop_sequences #14782
Conversation
The latest updates on your projects. Learn more about Vercel for Git ↗︎ 1 Ignored Deployment
|
Hi @hwchase17 ! Happy new year! I think this feature could be useful while providing a convenient layer of abstraction that centralizes the streaming logic. Who would be best suited for giving it a look? Many thanks in advance. |
Hi @baskaryan, @eyurtsev, @hwchase17. In a nutshell, this PR enables arbitrary |
@hwchase17 Hi! Noticed that you closed this PR. |
not intentional, we can't reopen at the moment but will do so as soon as we can :/ See #16796 |
Hi! @hwchase17 @baskaryan |
howdy! This is in the deprecated huggingface community integration - could you confirm this is still an issue in the partner package, and if so, submit an issue and/or PR against that? |
Introduces
StopSequenceAwareStreamer
that enhances streaming control by allowing end users to specify arbitrary stop sequences spanning multiple tokens, without knowledge of the tokenizer. This generalizes current functionality, adding capabilities without changing API. End users' current implementations will work the same while allowing longer stop sequences.Happy Friday <:)
This is still a draft in the sense that I would welcome your thoughts on: 1) where this logic would best be centralized. 2) which LLMs that we want to implement this for.
I've tested this out locally but will happily add unit tests for the components once we firm up the above two points!