[BUG/ERROR] Can only concatenate str (not "dict") to str

[j3ramy]

Hi,

I got an interesting bug/error (?) when I try to execute holmes ask command. I installed holmesgpt 0.7.2 via brew on my MacOS with Python version 3.9.6. The installation was successful, and I can execute the holmes version command.

I am using Ollama as the LLM backend (even if it may be buggy as your documentation says).

Everytime when I try to execute holmes ask "Which pod is not running?" --model=ollama_chat/llama3.2:1b then it runs into the following error:

in completion:2804
in get_ollama_response:366
in token_counter:1638

Failed to execute script 'holmes' due to unhandled exception!
TypeError: can only concatenate str (not "dict") to str

During handling of the above exception, another exception occurred:
in ask:281
in prompt_call:80
in call:122
in completion:148
in wrapper:960
in wrapper:849
in completion:3065
...

Is it because I am using Ollama or the Python version? Have you already experienced this error?

Thanks in advance!

[aantn]

Hi, I suspect this is a bug with LiteLLM (one of our dependencies) and how they implement Ollama support - maybe this bug? BerriAI/litellm#6958

Are you able to verify it doesn't happen with other models?

[j3ramy]

Hi, I suspect this is a bug with LiteLLM (one of our dependencies) and how they implement Ollama support - maybe this bug? BerriAI/litellm#6958

Are you able to verify it doesn't happen with other models?

Actually it happens with llama3.2:latest as well...

If I use gemma2:2b then this is the console output:

holmes ask "Why ist my cluster not running?" --model=ollama_chat/gemma2:2b  
User: Why ist my cluster not running?
Couldn't find model's name ollama_chat/gemma2:2b in litellm's model list, fallback to 128k tokens for              llm.py:140
max_input_tokens                                                                                                             
Couldn't find model's name ollama_chat/gemma2:2b in litellm's model list, fallback to 4096 tokens for              llm.py:174
max_output_tokens                                                                                                            
AI: {}

It's not finding the LLM even if it exists:

ollama list
NAME               ID              SIZE      MODIFIED   
llama3.2:1b        baf6a787fdff    1.3 GB    2 days ago    
gemma2:2b          8ccf136fdd52    1.6 GB    2 days ago    
llama3.2:latest    a80c4f17acd5    2.0 GB    2 d                                       

[aantn]

Just to update, this is not fixed in LiteLLM so it should be fixed in the next Holmes release after we updated LiteLLM.