LlamaIndex

Log inLog into community

Find answers from the community

Updated 2 months ago

Hi all, I am having an issue today with

Hi all, I am having an issue today with

·

Hi all, I am having an issue today with the OpenAI Agent using gpt-4-1106-preview, while yesterday everything worked fine, it looks as if today when the tool responds (the results are coming correctly in JSON from the external API) the Agent has a shorter time out and doesn't process the response from the tool. Everything works ok when the external API responds faster (and with lesser data).

L

c

23 comments

You can manually set the timeout in the LLM object -- I think the default is 60s?

llm = OpenAI(...., timeout=60.0)

ah this is a good suggestion, thanks a lot

I will keep you posted

I'm surprised 60s isn't long enough for a single response, but I guess these newer models are probably a lot slower right now

I will try and let you know, 60.0 is a long time

nope, it doesn't work....

llm = OpenAI(model="gpt-4-1106-preview", temperature=0.0, timeout=60.0)

Is it timing out after 60s?

Or can you paste some of the error? Curious what's actually happening lol

yes very interesting

Attachment

first query - the agent resets (or looks like) - no error but it asks me what I need

the second query the results are analyzed correctly

there are no errors

interestingly enough the agent on ChatGPT works for that same query

Attachment

Is this using OpenAIAssitant API? Or what kind of setup do you have?

super confused haha

so I have the same API that is used by a local agent created using Llama Index (OpenAI Agent) and ChatGPT

local agent is running on agent.wordlift.io

the response of the API, in some cases, might overflow GPT-3.5 but works with GPT-4

API takes 10s to respond

timeout=80.0 seems to work, I will try to move to Azure tomorrow and see if the response gets better, the futre is there now 🙂

Add a reply

Sign up and join the conversation on Discord