Assistant thread throwing AceInternalException

I am developing an assistant and there are times when I will encounter an AceInternalException when asking the assistant to process data attached to the assistant. Once this happens all use of code_interpreter results in a failure and the thread has to be destroyed/recreated in order to continue work. As you can imagine this is a major issue as I lose all context I’ve built in the thread upon experiencing these failures.

Is OpenAI aware that this is still an issue? I found a similar issue posted in Nov’23 that didn’t have a resolution other than ‘wait it out’. Does anyone know of a way to reset or reboot the code_interpreter to get around this issue, or is starting a new thread the only way to accomplish this?

EDIT: The problem seems to happen either at random (e.g., in the middle of an active thread) or after a period of inactivity on the thread. Is there an internal timeout set on code_interpreter instances attached to an assistant thread? If so, is there a way to inspect this? Having the assistant randomly crap-out is a horrible UX.

6 Likes

Hi,
I am also getting AceInternalExceptions after sometime when working with uploaded datasets, initial responses are fine but then I am getting crap-out and hallucinated answers…

It seem to be a bug exiting since Nov 2023.
To prevent the model from false answers I am adding the following instruction:

As a general instruction, do not answer in cases of issues connecting to the data structures or failure in retrieving the requested information.
Just write that there was a problem and add details about it if you have them.

Should be added as system instruction.

I am getting the same issue after generating a single word document(docx) and I suspect it is related to file streams not being closed properly.

This bug ruins the context of the conversation and the only way for me to get around it is to start a new conversation.

I can ask the GPT to generate a prompt to transfer the context to a new converastion.

As soon as the next document is generated I find myself in the same situation again.

4 posts were split to a new topic: Can’t access files in code interpreter after a period of inactivity