How to use Ollama models in Rasa CALM?

Rasa Pro CALM
1

Hello, I am trying to use models served through Ollama into Rasa CALM? I could not find it anywhere in the documentation, any leads or tips would be helpful

2

Hi Vishni,

The easiest way I’ve found to use Ollama (presuming you’re running the Ollama server locally) is to make use of its OpenAI-compatible entry point.

Your config would look similar to:

pipeline:
- name: LLMCommandGenerator
  llm:
    model: "wizardlm2:7b"
    max_tokens: 20
    type: openai
    openai_api_base: http://localhost:11434/v1
    openai_api_key: foobar

And, of course, you can replace the value for model with whichever model you have downloaded to your Ollama installation.

Hope this helps!

3

Hi Chris, Thanks for the quick response! Sure, this is of great help, will try it out.

4 
2024-05-13 17:28:19 WARNING  langchain.llms.base  - Retrying langchain.llms.openai.completion_with_retry.<locals>._completion_with_retry in 8.0 seconds as it raised Timeout: Request timed out: HTTPConnectionPool(host='localhost', port=11434): Read timed out. (read timeout=7).
2024-05-13 17:28:34 WARNING  langchain.llms.base  - Retrying langchain.llms.openai.completion_with_retry.<locals>._completion_with_retry in 10.0 seconds as it raised Timeout: Request timed out: HTTPConnectionPool(host='localhost', port=11434): Read timed out. (read timeout=7).
2024-05-13 17:28:51 ERROR    rasa.utils.log_utils  - [error    ] llm_command_generator.llm.error error=Timeout(message="Request timed out: HTTPConnectionPool(host='localhost', port=11434): Read timed out. (read timeout=7)", http_status=None, request_id=None)

I tried to use Ollama as instructed, and it threw me this error. I suspect its the same issue discussed here and according to the status of the issue, doesn’t seem to be getting a fix anytime soon. So can anyone suggest alternatives for running local models?

5

Hi Vivek,

Are you able to access your locally running Ollama server with curl? e.g.

curl http://localhost:11434/api/generate -d '{ "model": "phi3", "prompt": "Why is the sky blue?", "stream": false }'

6

Yes, Ollama is accessible via curl and the python openAI api interface, but doesn’t seem to work when called via Rasa. The config is

pipeline:
- name: LLMCommandGenerator
  llm:
    model: "llama3:latest"
    type: openai
    openai_api_base: http://localhost:11434/v1
    openai_api_key: foobar
7

@MannavaVivek did you found any solution?

8

With the upgraded version of Rasa(mine is 3.9.3), you can directly call the model

type: ollama model: “llama3.1:8b”

9

Did the update solve it?