annoying bug

[2312adas]

error> non-retryable provider error: Model 'custom/kimi-k2.7-code' rejected the request (400). This usually
means a parameter the model no longer supports (e.g. temperature on Claude Opus 4.7). Underlying: Error code:
400 - {'error': {'message': '/chat/completions: Invalid model name passed in model=custom/kimi-k2.7-code. Call
/v1/models to view available models for your key.', 'type': 'None', 'param': 'None', 'code': '400',
'[REDACTED]': {'error': '/chat/completions: Invalid model name passed in model=custom/kimi-k2.7-code. Call
/v1/models to view available models for your key.'}}}

──────────────────────────────────────────────────────────────────────────────────────────────────────────────
›

[PurpleCHOIms]

This is a configuration issue rather than a bug in Decepticon. The underlying error comes straight from LiteLLM:

Invalid model name passed in model=custom/kimi-k2.7-code.
Call `/v1/models` to view available models for your key.

custom/kimi-k2.7-code isn't a model your LiteLLM proxy knows about — it has to be registered in config/litellm.yaml (a model_name entry with litellm_params pointing at the real provider endpoint + key) and match a constant in decepticon/llm/models.py. Right now the proxy has no route for that name, so it 400s.

To fix:

1. Add the model to config/litellm.yaml (or your override config) with a valid litellm_params.model for whichever provider actually serves kimi-k2.
2. Restart the litellm container so it reloads the config.
3. Verify it resolves: curl http://localhost:4000/v1/models should list it.

If you were expecting that name to work out of the box, let me know where you configured it (env var / config file) and I'll point at the exact spot. Marking needs-info — happy to reopen/dig in with your config. That said, one fair takeaway: the error could name which config file to edit, so I'll consider a small message improvement.

[cesar2926-rgb]

Run.exe