Request | Make it possible to configure a remote Ollama host/baseURL

[aweijnitz]

Currently the Ollama class assumes Ollama is running on localhost. This is an unnecessary limitation.
The config object is not passed down to the "native wrapper" although it looks like it in the code and thus, it defaults to localhost and port.

As a developer, I would like to supply a host and port to be used when creating my Ollama instance.

Example of intended use (see config object)

Settings.llm = new Ollama({ 
    model: models[modelIndex],
    temperature: 0.75,
    config: {
      host: 'http://my.own.internal.hostname.de:${CUSTOM_PORT}/'
    }
  })

The "native wrapper" class where the config object never arrives:

LlamaIndexTS/packages/llamaindex/src/internal/deps/ollama.js

Line 274 in 41fe871

host: "",

[gcapozzo]

Set this in the .env file:
OLLAMA_BASE_URL=http://IP:PORT

[himself65]

I'm about to remove the wrapper class