Open JaloWalt opened 2 weeks ago
When testing this issue with the newest autogen version: 0.0.56rc12
the behavior was as follows:
Creating a new model: it saved three times each new created model, but nothing new became visible in Models view.
After that any new models created did come visible in Models view.
In creating new models (after the first three [see above]) a new selection was presented: see attached picture; pre-choosing-model-type-2024-06-11_120029.jpg
Having a pre-choosing of model type between: OpenAI, Azure OpenAI and Gemini
However none of those worked (tested with the test-button) with the actual fireworks.ai base-url: https://api.fireworks.ai/inference/v1/chat/completions
Therefore, the informed bug still remains.
I had a similar bug open: https://github.com/microsoft/autogen/issues/2889
Could be more of a documentation/usability issue as both these env variables need to be set:
export OPENAI_API_BASE="http://192.168.1.64:5000/v1"
export OPENAI_BASE_URL="http://192.168.1.64:5000/v1"
If these need to be set as environment variables:
OPENAI_API_BASE="http://192.168.1.64:5000/v1" OPENAI_BASE_URL="http://192.168.1.64:5000/v1"
then (A) (where does it come from?) /what is this: 192.168.1.64
Asking because it seems to be some internal address?
then (B) what is this: 5000
Surely not some random port address?
In summary: I am trying to understand the connection of those (A) and (B) to some real system dependent item.
Describe the bug
For example with fireworks.ai model: FireFunction V1
Ref: https://fireworks.ai/models/fireworks/firefunction-v1
The base url is informed as: https://api.fireworks.ai/inference/v1/chat/completions
That would then result to the folllowing settings in autogen studio model specification:
Model name: accounts/fireworks/models/firefunction-v1
Api Key: your-fireworks-ai-secret-api-key
Base url: https://api.fireworks.ai/inference/v1/chat/completions
Api Type: empty
Api version: empty
Description: fireworks firefunction-v1
When testing this setup with "test-model" button the result is: => Error occurred while testing model: Error code: 404 - {'detail': 'Not Found'}
The log shows: File "\site-packages\openai_base_client.py", line 1020, in _request
raise self._make_status_error_from_response(err.response) from None
openai.NotFoundError: Error code: 404 - {'detail': 'Not Found'}
However when the [Base url] is changed to: https://api.fireworks.ai/inference/v1
Then the result (with the same test) is: Model tested successfully
Therefore the conclusion is: Autogen Studio App does not respect the base url setting given by api-inference providers.
This is a serious BUG.
Steps to reproduce
For example with fireworks.ai model: FireFunction V1
Ref: https://fireworks.ai/models/fireworks/firefunction-v1
The base url is informed as: https://api.fireworks.ai/inference/v1/chat/completions
That would then result to the folllowing settings in autogen studio model specification:
Model name: accounts/fireworks/models/firefunction-v1
Api Key: your-fireworks-ai-secret-api-key
Base url: https://api.fireworks.ai/inference/v1/chat/completions
Api Type: empty
Api version: empty
Description: fireworks firefunction-v1
When testing this setup with "test-model" button the result is: => Error occurred while testing model: Error code: 404 - {'detail': 'Not Found'}
The log shows: File "\site-packages\openai_base_client.py", line 1020, in _request
raise self._make_status_error_from_response(err.response) from None
openai.NotFoundError: Error code: 404 - {'detail': 'Not Found'}
However when the [Base url] is changed to: https://api.fireworks.ai/inference/v1
Then the result (with the same test) is: Model tested successfully
Therefore the conclusion is: Autogen Studio App does not respect the base url setting given by api-inference providers.
This is a serious BUG.
Model Used
accounts/fireworks/models/firefunction-v1
Expected Behavior
The folllowing settings in autogen studio model specification should have worked:
Model name: accounts/fireworks/models/firefunction-v1
Api Key: your-fireworks-ai-secret-api-key
Base url: https://api.fireworks.ai/inference/v1/chat/completions
Api Type: empty
Api version: empty
Description: fireworks firefunction-v1
When testing this setup with "test-model" button the result should have been: => Model tested successfully
Screenshots and logs
File "\site-packages\openai_base_client.py", line 1020, in _request
raise self._make_status_error_from_response(err.response) from None
openai.NotFoundError: Error code: 404 - {'detail': 'Not Found'}
Additional Information
Autogen Studio version: v0.0.56 puautogen version: pyautogen-0.2.27-py3-none-any.whl (as informed in build log) Python version: Python 3.12.3