using local ollama models returns a json structure instead of executing them

[jwcloud365]

Describe the bug

Today I installed openinterpreter and want to use it locally.

But when I'm using a local ollama model, I get some json structure as an output:

Loading qwen2.5-coder:32b...

Model loaded.

Open Interpreter will require approval before running code.

Use interpreter -y to bypass this.

Press CTRL-C to exit.

list all files in my Images directory

{"name": "execute", "arguments":{"language": "shell", "code": "dir %USERPROFILE%\Pictures"}}

browse to cnn.com

  {"name": "execute", "arguments":{"language": "python", "code":
  "computer.browser.setup(headless=False)\ncomputer.browser.go_to_url('https://www.cnn.com')"}}

But when i use an openai model, I get this output, as expected:

list all files in my Images directory

ls ~/Images

Would you like to run this code? (y/n)

If you need further actions or details on any specific file, please let me know!

browse to cnn.com

  computer.browser.go_to_url('https://www.cnn.com')


  Would you like to run this code? (y/n)

Reproduce

interpreter

list all files in my Images directory

{"name": "execute", "arguments":{"language": "shell", "code": "dir %USERPROFILE%\Pictures"}}

browse to cnn.com

  {"name": "execute", "arguments":{"language": "python", "code":
  "computer.browser.setup(headless=False)\ncomputer.browser.go_to_url('https://www.cnn.com')"}}

Expected behavior

  computer.browser.go_to_url('https://www.cnn.com')


  Would you like to run this code? (y/n)

Screenshots

I would expect

instead of

{"name": "execute", "arguments":{"language": "python", "code":
"computer.browser.setup(headless=False)\ncomputer.browser.go_to_url('https://www.cnn.com')"}}

Open Interpreter version

Version: 0.4.3

Python version

Python 3.11.11

Operating System name and version

W11

Additional context

No response

[thomahn3]

Same here using phi, llama3.2 and mistral on macos 15.2.

[nainglinwai1]

For those problems, I found the same error after first try. I am using the python 3.10. I can run with llama3.2:latest model and open-interptreter local tutorial. It's working now.

[66Ton99]

The same, but it works OK with qwen

[nainglinwai1]

I found the solution to make it works.
First you need to export OLLAMA_MODEL with this
export OLLAMA_MODEL=llama3.2:latest
export OLLAMA_API_BASE=http://localhost:11434/v1

Run these before you run the code.

[thomahn3]

I found the solution to make it works. First you need to export OLLAMA_MODEL with this export OLLAMA_MODEL=llama3.2:latest export OLLAMA_API_BASE=http://localhost:11434/v1

Run these before you run the code.

I ran this code before running interpreter and now I get the following error: 16:39:34 - LiteLLM:ERROR: utils.py:1873 - Model not found or error in checking vision support. You passed model=llama3.2:latest, custom_llm_provider=ollama. Error: OllamaError: Error getting model info for llama3.2:latest. Set Ollama API Base via 'OLLAMA_API_BASE' environment variable. Error: Client error '404 Not Found' for url 'http://localhost:11434/v1/api/show/api/show'

[nainglinwai1]

I found the solution to make it works. First you need to export OLLAMA_MODEL with this export OLLAMA_MODEL=llama3.2:latest export OLLAMA_API_BASE=http://localhost:11434/v1
Run these before you run the code.

I ran this code before running interpreter and now I get the following error: 16:39:34 - LiteLLM:ERROR: utils.py:1873 - Model not found or error in checking vision support. You passed model=llama3.2:latest, custom_llm_provider=ollama. Error: OllamaError: Error getting model info for llama3.2:latest. Set Ollama API Base via 'OLLAMA_API_BASE' environment variable. Error: Client error '404 Not Found' for url 'http://localhost:11434/v1/api/show/api/show'

you need to pull the model you want first. For me I am using llama3.2:latest. You should add the model you want to use

[ahabiba]

Disable Function Calling: Some models, especially those from Ollama, are designed to return structured outputs for function execution. To handle this, you can inform Open Interpreter that the model doesn't support function calling by using the --no-llm_supports_functions flag. This approach has been suggested in Issue #1412.

Example Command:

interpreter --model ollama/mistral:7b-instruct --no-llm_supports_functions