Support for server APIs of apps specialised in local LLMs

[ThiloteE]

Solves support for

• local LLMs
• hardware
• gpu acceleration
• custom and more LLM architectures

Describe the solution you'd like

Support for applications that adhere to the OpenAI API.
OpenAI API has become an inofficial standard.

Progress:

• Llama.cpp
  • JabRef documentation
  • Documentation: https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md
• GPT4All
  • Support GPT4All Server API #11870
  • Send embeddings to GPT4All's Local API Server #12114
  • system prompt is ignored in API request nomic-ai/gpt4all#1855
  • Chat UI Server: Support listening on address other than localhost nomic-ai/gpt4all#1304
  • Local server not remembering previous messages nomic-ai/gpt4all#2602
  • Convert API server to use OpenAPI spec nomic-ai/gpt4all#3113
  • JabRef documentation
  • Documentation: https://github.com/nomic-ai/gpt4all/wiki/Local-API-Server
• LMStudio
  • JabRef documentation
  • Documentation: https://lmstudio.ai/docs/basics/server
• Ollama
  • JabRef documentation: https://docs.jabref.org/ai/local-llm#step-by-step-guide-for-ollama
  • Documentation: https://github.com/ollama/ollama/blob/main/docs/api.md
• Jan
  • JabRef documentation
  • Documentation: https://jan.ai/integrations/coding/continue-dev#step-2-enable-the-jan-api-server. I think they are using https://github.com/janhq/cortex.cpp?tab=readme-ov-file#overview
• KoboldCPP
  • JabRef documentation
  • Documentation: https://lite.koboldai.net/koboldcpp_api
• Llamafile
  • JabRef documentation
  • Documentation: https://github.com/Mozilla-Ocho/llamafile?tab=readme-ov-file#quickstart

Most of the applications listed here are a wrapper around llama.cpp, though they all have their unique strenghts and weaknesses. Except for LMStudio, they are all open source.

Related issues:

InAnYan#71

Notes:

This issue here is purely about LLMs. Not embedding models. For embedding models see InAnYan#85 (comment).