Inconsistent Responses with Azure OpenAI Models in Dify 1.0.1

[hosushi]

Self Checks

• This is only for bug report, if you would like to ask a question, please head to Discussions.
• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:）
• Please do not modify this template :) and fill in all the required fields.

Dify version

v1.0.1

Cloud or Self Hosted

Self Hosted (Docker)

Steps to reproduce

• Update to Dify 1.0.1.
• Use the Azure OpenAI model for an image-to-text extraction task.
• Upload an image and execute the task.
• Observe that the response returned is inconsistent with the content of the uploaded image.

✔️ Expected Behavior

• The responses should accurately reflect the content of the uploaded image.
• For example, in v0.14.3 using Azure OpenAI@GPT4o, although the accuracy was somewhat low, the responses were aligned with the image content.

❌ Actual Behavior

• In Dify 1.0.1:
  • Azure OpenAI@GPT4o: The responses do not match the image content.
  • Azure OpenAI@o1: The responses do not match the image content.
• Additionally, when using the o1 model in the Azure OpenAI Playground, the expected highly accurate responses are returned without any issues.
• In rare cases, even when using the same image for all tests, the Azure OpenAI-related models in v1.0.1 sometimes return responses indicating that the resolution is too low and the text is unrecognizable.

[hosushi]

I have confirmed that the same phenomenon occurs in Dify 1.0.0, so I am adding this note.