Unable to create new Provider if model_id_key="model_id"

[michaelchia]

Problem

• I am unable to create my own provider if the model_id_key for my langchain LLM object is "model_id".
  • the "model_id" kwarg would be repeated twice in super().__init__ in L122
    

    jupyter-ai/packages/jupyter-ai-magics/jupyter_ai_magics/providers.py

    Lines 111 to 122 in 96465d1

    	def __init__(self, *args, **kwargs):
    	try:
    	assert kwargs["model_id"]
    	except:
    	raise AssertionError(
    	"model_id was not specified. Please specify it as a keyword argument."
    	)
    	model_kwargs = {}
    	model_kwargs[self.__class__.model_id_key] = kwargs["model_id"]
    	super().__init__(*args, **kwargs, **model_kwargs)

• Not sure if this problem exists for BaseEmbeddingProvider as it seems to be implemented differently.

Proposed Solution

change L120 to:

if self.__class__.model_id_key != "model_id":
    model_kwargs[self.__class__.model_id_key] = kwargs["model_id"]

[dlqqq]

@michaelchia I wouldn't expect that to raise an exception, since unpacking two dictionaries with shared keys does not; it only results in the second dictionary clobbering the first:

>>> a = { "key": 1 }
>>> b = { "key": 2 }
>>> c = { **a, **b }
>>> c
{'key': 2}

If this is raising an exception, could you post the exception info here? If not, I'm not clear on what the issue is.

[michaelchia]

Repeated keys are allowed in curly brackets but not for functions. e.g. {"key": 1, "key": 2} won't throw errors but something like dict(key=1, key=2) (or dict(**a, **b) as per your example) will throw an exception (TypeError: dict() got multiple values for keyword argument 'key'). if it were super().__init__(*args, **{**kwargs, **model_kwargs}) it would've been fine.

[3coins]

@michaelchia
I have encountered this issue as well, and attempted to resolve this in PR #263.