LiteLLM Example in mistral docs wrong

[ishaan-jaff]

Operating System

MacOS

Version Information

not relevant

Steps to reproduce

https://github.com/Azure/azureml-examples/blob/main/sdk/python/foundation-models/mistral/litellm.ipynb
@santiagxf Thanks for showing an example with litellm, but the docs are wrong

Here's how to use litellm with mistral
https://docs.litellm.ai/docs/providers/mistral

from litellm import completion
import os

os.environ['MISTRAL_API_KEY'] = ""
response = completion(
    model="mistral/mistral-tiny", 
    api_base="your-api-base",
    messages=[
       {"role": "user", "content": "hello from litellm"}
   ],
)
print(response)

Expected behavior

Actual behavior

Addition information

No response

[santiagxf]

@ishaan-jaff, the example is correct. You are sharing an example about how to use Mistral's inference platform, but this is Azure AI. Does it make sense?

[ishaan-jaff]

any reason why you could not use it like this ? @santiagxf ? (i'm the maintainer of litellm)

This looks a lot easier to me and it can go to to an Azure AI endpoint

from litellm import completion
import os

os.environ['MISTRAL_API_KEY'] = ""
response = completion(
    model="mistral/mistral-tiny", 
    api_base="your-api-base",
    messages=[
       {"role": "user", "content": "hello from litellm"}
   ],
)
print(response)

[ishaan-jaff]

Hi @santiagxf I just deployed on Azure AI studio and I was able to run inference with this code

If possible can we update the python notebook with the following code ? It uses the standard format on litellm docs: https://docs.litellm.ai/docs/providers/azure_ai

Happy to make a PR for this too

from litellm import completion
import os

response = completion(
    model="mistral/Mistral-large-dfgfj", 
    api_base="https://Mistral-large-dfgfj-serverless.eastus2.inference.ai.azure.com/v1",
    api_key = "JGbKodRcTp****"
    messages=[
       {"role": "user", "content": "hello from litellm"}
   ],
)
print(response)

[santiagxf]

We tried your example, but it looks the api_base is hardcorded somewhere. Hence you get an access denied error because the token is going to the wrong API. You will get this error (I got this by setting litellm.set_verbose=True).

POST Request Sent from LiteLLM:
curl -X POST \
https://api.mistral.ai/v1/ \
-d '{'model': 'Mistral-large-dfgfj', 'messages': [{'role': 'user', 'content': 'hello from litellm'}], 'extra_body': {}}'

Is this something you can fix? @ishaan-jaff

[ishaan-jaff]

Yes we fixed this today: BerriAI/litellm#2216 @santiagxf

Thanks for raising this

[santiagxf]

I just had a look and notice that it requires v1 at the end of the URL. mistral-python, the official client, doesn't require it. I think we should fix that to work in the same way.

https://github.com/BerriAI/litellm/blob/e56cc26e18b83aa39617b7d2cf8ce3c1e3dfee87/litellm/utils.py#L4931

Also, I spotted this line:

https://github.com/BerriAI/litellm/blob/e56cc26e18b83aa39617b7d2cf8ce3c1e3dfee87/litellm/utils.py#L4932

Does it means api_key is being ignored if passed as argument?

[ishaan-jaff]

Does it means api_key is being ignored if passed as argument?

Nope, that line ensure we use MISTRAL_API_KEY from the env

think we should fix that to work in the same way.

Do you mean litellm the package should append a /v1 to Azure AI studio api_base if the user does not add it ?

[ishaan-jaff]

Does it means api_key is being ignored if passed as argument?

I think you're right about this actually, going to take a look at this

Summarizing next steps from litellm

• Stop requiring /v1 for mistral azure ai studio
• ensure api key is not ignored when passed for mistral/

Does this sound good @santiagxf ? Should I make a PR to this repo once this is fixed

[santiagxf]

It sounds good to me. The URL thing is not a big deal, but I think it would be nice to use the same approach the official client uses.

Thanks for taking care of this! Once the new version of the library is out we can update this doc. Feel free to use this issue. I will keep it opened so I can remember to do it.

[hsleiman1]

Hi,
Why do we need to mention the api_base and api_key, which are not needed for the other models? It would be great if they are also read from the env variables.
Thank you!

[ishaan-jaff]

@hsleiman1 noted - will add this as an improvement too

[ishaan-jaff]

Tracking this here @hsleiman1 BerriAI/litellm#2237

[ishaan-jaff]

Fixes are here : BerriAI/litellm#2247 @hsleiman1 @santiagxf, will update once a new release is out

[santiagxf]

@ishaan-jaff we still see the same issues happening. I added the comment in the issue you created on your repo:

BerriAI/litellm#2237 (comment)