The community members are discussing the standard way to format the response sent back, particularly regarding the removal of <|im_end|> tags. One community member is using AzureOpenAI and is unsure if the quality of the documents they are feeding for embedding is affecting the response. Another community member has switched from using ChatGPT 3.5 deployment to AzureChatOpenAI and has seen better results. There is some confusion around the comparison between Azure GPT 3.5 and OpenAI GPT 3.5, with the community member clarifying that they have only used AzureOpenAI and AzureChatOpenAI, and never directly used OpenAI.
Sorry for the confusion, I used AzureOpenAI llm then switched to AzureChatOpenAI impl with a chatgpt3.5-turbo deployment and got better results. Never used OpenAi directly