Added API Proxy functionality to API LM #68
Conversation
AakritiKinra
changed the title
There was a problem hiding this comment.
Thanks @AakritiKinra !
But I think this changes the signature of APIBasedLM.generate() to be different than that of the parent LanguageModel.generate(). We should keep the signature of parent and child abstract functions the same.
Instead, I think you can specify api_base in the constructor of APIBasedLM.
|
I have added the api_base to the constructor. The signature of the two generate functions is still different because of the completion function. I am passing the message which is the prompt for which the response is generated. In the generate function of the LanguageModel class, there is no message parameter. The chat_generate function of both classes has messages as a parameter. |
|
it looks like the modified version is not passing tests. We’ll have to make sure that the “generate” interface is the same for all LMs |
|
Thanks! |
* added base_url * Updated function descriptions * Added api_base to the constructor * matched structure with lm class Pull latest changes
Description
Added the api_base parameter to the completion() and batch_completion() functions to use the API Proxy.
References