Open rmax opened 1 month ago
Comment from duplicate #2763 https://discord.com/channels/1108621136150929458/1300903286060220469
Can't currently do fully custom endpoints that aren't compatible with e.g. openai, ollama without adding a provider in the codebase. e.g. company.com/embeddings, company.com/conversation
baseUrl and apiUrl are nice but assume the endpoints match with e.g. ollama or openai
Solution Add chatUrl, embeddingsUrl etc. or some other paradigm that would allow this
Falls under meta llm-info + config #2769
Validations
Problem
I’m trying to use CodeLlama deployed on databricks but the endpoint is /invocations rather than /completions, using OpenAI-compatible provided.
Solution
It would be useful if we could override the completion endpoint by configuration.