aider-qwq-coder example bug

[gilbrotheraway]

i am following the example here> https://github.com/mostlygeek/llama-swap/tree/main/examples/aider-qwq-coder

but keep getting this error:

**litellm**.AuthenticationError: AuthenticationError: OpenAIException - The api_key client option must 
be set either by passing api_key to the client or by setting the OPENAI_API_KEY environment variable
The API provider is not able to authenticate you. Check your API key.

afaik llama-swap has no api key

also am confused why sampling parameters in the example diverge for aider.model.settings and llama-swap config files??

[aa956]

afaik llama-swap has no api key

Yes, so you can use any (not empty) string.

I'm using the following .aider.conf.yml in project directory:

model: openai/qwen3-32b-q4
## Use architect edit format for the main chat
architect: true
editor-model: openai/qwen2.5-coder-32b-q4
model-settings-file: .aider.model.settings.yml
openai-api-key: anything-at-all-here
openai-api-base: http://localhost:8000/v1
model-metadata-file: .aider.model.metadata.json

Result:

$ aider
───────────────────────────────────
Aider v0.83.2
Main model: openai/qwen3-32b-q4 with architect edit format
Editor model: openai/qwen2.5-coder-32b-q4 with editor-diff edit format
Weak model: openai/qwen2.5-coder-32b-q4
Git repo: .git with 6 files
Repo-map: using 4096.0 tokens, auto refresh
───────────────────────────────────
architect>

[gilbrotheraway]

it was a typo sorry for the noise, btw i couldn't get the editor model in no_think mode no matter what i set in llamaswap, aider.model settings or editor-whole config file

[mostlygeek]

@Fuckingnameless

try —reasoning-budget 0 for the editor model. ggml-org/llama.cpp#13771

[gilbrotheraway]

@Fuckingnameless

try —reasoning-budget 0 for the editor model. ggml-org/llama.cpp#13771

on aider.model settings file?
I'm on ik_llama.cpp since it runs 100% faster on my dual socket xeon

[aa956]

@Fuckingnameless
try —reasoning-budget 0 for the editor model. ggml-org/llama.cpp#13771

on aider.model settings file? I'm on ik_llama.cpp since it runs 100% faster on my dual socket xeon

No, it's a server parameter.

ik_llama does not have it.

So the simplest way to disable thinking with ik_llama server seems to be adding /no_think to every prompt.

Look at aider's conventions for that.

But then your model should support that, like Qwen 3

[gilbrotheraway]

@Fuckingnameless
try —reasoning-budget 0 for the editor model. ggml-org/llama.cpp#13771

on aider.model settings file? I'm on ik_llama.cpp since it runs 100% faster on my dual socket xeon

No, it's a server parameter.

ik_llama does not have it.

So the simplest way to disable thinking with ik_llama server seems to be adding /no_think to every prompt.

Look at aider's conventions for that.

But then your model should support that, like Qwen 3

not working here neither with the system prompt, but it's time i moved to vllm/tabby

thanks for the feedback people