Why ChatOpenAI is slow response in sometimes

[kientv]

Columns: Question, Answer, Datetime, Delay, Tokens and last colum is first token:

I don't know why it's slow. Is it problem of OpenAI (GPT-3.5-turbo-0125) or Langchain and how to improve it?
Becasue facebook app reviewer test and they does not approve our chatbot on messenger.

We determined that the use case was valid. However, we were unable to verify the requested permission. During the app review, we did not receive an automated response when we attempted to message your bot. Bots should respond to all inputs within 30 seconds to prevent a poor experience.

[maximeperrindev]

Hi @kientv
Since when do you have this problem ?
I've heard OpenAI had some issues with gpt-3.5 models today

[kientv]

Thanks @maximeperrindev
This problem happen today with GPT-3.5-turbo-0125 (in start time column is GMT+7)
I see:

API Status: Increased error rates and latency on GPT-3.5-Turbo and ChatGPT
Between 3:35AM – 4:50AM PST we observed elevated error rates and latency on GPT-3.5-Turbo, GPT-3.5-Turbo-1106, and some ChatGPT models. We have implemented a fix and are monitoring. Learn more

https://status.openai.com

Seem that their problem (PST) is not match with our error times (GMT+7).

[maximeperrindev]

@kientv
Yeah you're right. Then here are some suggestions you could do to find the problem :

• Try benchmarking with more tries (this could help to check if it's code-related or provider-related
• Change model to GPT-4
• Add time breakpoints at different places in your code to see what is taking too much time