Memory?

[gorilla-san]

First of all, this is fantastic! Got it working in like 15 mins. Pretty cool stuff.

But for some reason the bot doesn't remember previous back-and-forths.

Is memory buffer implemented or does it not have it?

I'll try figuring that out myself, too (I tried in like 10 minutes, but gotta go sleep now and it's gonna take me hours to figure it out as I'm a newbie), but thought it'd save me some time if somebody tells me if it's even implemented or not :-)

Thank you!

[stefansms]

I believe I am experiencing the same problem. The model behaves well in the first question. However, in subsequent questions it gets stuck with the sources from the first question.

Even if the second question is from the same document, but on another subject, it cannot follow the conversation.

Does anyone else have this same problem?

[danukito899]

Sane here, please help

[CharleoY]

This is because when the length grows, the model will decrease the attention to the contextual information. You can try in the chatGPT's web UI and use the max length it supports. You can see by after 2 questions it will forget the previous information since the length of input reaches the limitation. Similar here, how can the model remember the information in the previous conversation? Because each of your input is like an array that contains the history message and self-append with the new message.

[stefansms]

I don't know if it is possible. But maybe we could use chatgpt to summarize the context to append in next question, keeping conversation flowing for at least 2 or 3 questions?

[dosubot[bot]]

Hi, @gorilla-san! I'm Dosu, and I'm helping the gpt4-pdf-chatbot-langchain team manage their backlog. I wanted to let you know that we are marking this issue as stale.

From what I understand, the issue is about the bot not remembering previous conversations. Other users have also reported experiencing the same problem and have suggested some potential solutions. One user mentioned that the model's attention to contextual information decreases as the conversation length grows. Another user suggested using chatGPT to summarize the context and append it to the next question to keep the conversation flowing.

Before we close this issue, we wanted to check with you if it is still relevant to the latest version of the gpt4-pdf-chatbot-langchain repository. If it is, please let us know by commenting on the issue. Otherwise, feel free to close the issue yourself, or it will be automatically closed in 7 days.

Thank you for your understanding and contribution to the gpt4-pdf-chatbot-langchain project!