✨ [Tasks] Add typings for the TGI / OpenAI chat API #480
Conversation
SBrandeis
force-pushed
the
text-generation-messages-api
branch
from
ad00ad5 to
537e06b
Compare
SBrandeis
changed the title
There was a problem hiding this comment.
are the types very different from the "classic" mode?
(if not, maybe we could embed them in the same task?)
There was a problem hiding this comment.
also not a huge fan of the name text-generation-conversational, would prefer text-generation-messages-api in that case
There was a problem hiding this comment.
are the types very different from the "classic" mode?
Sadly looks quite different from our text-generation. Our types are all based on a dict with inputs and parameters. Here parameters seem to be at the root level instead of a level below. And inputs (simple string) is replaced by the messages list of items.
Merging both specs doesn't seem to have conflicting keys but would result in a quite ugly definition no?
|
Closing in favor of #537 |
Ping @julien-c @Wauplin @osanseviero
parameters from https://huggingface.github.io/text-generation-inference/#/Text%20Generation%20Inference/chat_completions