Open limusen75 opened 11 hours ago
@limusen75 Hello there! I'm here to help you with any bugs, questions, or contributions. Let's work together to solve the issue you're facing.
The discrepancy you're observing between the documentation and the actual behavior of the OpenAI API-Compatible settings in Dify might be due to outdated documentation or a frontend issue. The current implementation in Dify does support function calling, as indicated by the presence of options for "function_call" and "tool_call" in the settings. When function calling is enabled, Dify should handle tools differently compared to when it is disabled.
If the documentation states that function calling is not supported, it may need updating to reflect the current capabilities of the platform. On the other hand, if the frontend is not behaving as expected (e.g., not sending tools when function calling is enabled), it could be a bug that needs addressing.
To resolve this, it would be beneficial to verify the documentation's accuracy and ensure that the frontend implementation aligns with the backend capabilities. If the issue persists, it might be worth reporting it for further investigation by the maintainers.
To continue talking to Dosu, mention @dosu.
@crazywoola I'm currently using the OpenAI API-Compatible configuration model llama3.1-70b-Instruct, with function calling enabled. However, when I send a request to the model using Dify, the request does not include the 'tools' parameter. Here are some related test screenshots:
Self Checks
Provide a description of requested docs changes
In the model provider list, it is written that OpenAI API-Compatible does not support function calling. However, in the OpenAI API-Compatible settings, there is an option for function calling, and I found that when I select to support function calling, Dify does not send the tools to the model during debugging. On the other hand, when I select not to support function calling, Dify sends the tools to the model in a React way during debugging. Could you please clarify whether the function calling option in OpenAI API-Compatible is a frontend bug or an outdated documentation issue in the model provider list?