Original Discord Post by darkladen | 2024-06-27 09:52:42
Hi, is there any way for the LLM to respond consistently with the documents loaded in the knowledge bank? We are putting together a product for several clients, a kind of totem of attention and in the first instance but not least, is that it responds clearly and concisely according to the information that has been uploaded. For example, if I ask him to recommend a place to eat and he already has information from a list of stores loaded, many times he answers “sure” or “yes I can” but says nothing more… it’s as if he keeps thinking but never says the answer. In the end I have to insist several times and sometimes he answers me with information that is not in the documents.
What I need is for the LLM model to respond faithfully to the information provided in the documents.
I have been working with llm’s like ChatGPT for a long time and when I load information in documents it always responds with this information and does not invent or forget part of the information.
We are creating a product for several customers but this issue is complicating our progress. I have only uploaded 3 documents and they are well reduced with information in each one.
Can anyone provide a proven way that actually works?
Is there any way to activate something so that the LLM always responds with the information loaded in its knowledge bank ?
In a previous ticket we had already partly solved the problem of it being able to recognize the stores but now it is a problem of memory loss or even mixing information and even inventor of it that does not exist in the documentation.
Hi, Any idea what problems are occurring ?
Now my avatar does not generate an output audio, I have already changed the voice and the problem persists.
I’m getting these error messages:
ConvaiGRPCLog: Warning: OnStreamFinish: Status:Not Ok | Debug Log: | Error message:Unable to generate audio response. Please verify that you have a valid voice set for the character. | Error Details:cUnable to generate audio response. Please verify that you have a valid voice set for the character. | Error Code:2 | Character ID:affe8d52-2d94-11ef-9ee4-42010a7be00e | Session ID:1ee9cca6069ee46afa304ed6ea699bdb
ConvaiChatbotComponentLog: Warning: UConvaiChatbotComponent Get Response Failed! | Character ID : affe8d52-2d94-11ef-9ee4-42010a7be00e | Session ID : 1ee9cca6069ee46afa304ed6ea699bdb
Is there anyone who can help with these problems?
We are having a lot of problems with the audio generation related to the error I have left before but I see that no one responds.
How long do I have to wait for a response and know that they are looking for a solution ?
It is urgent to know if we will have support and response to the problem.
OK thanks, I wouldn’t spam other channels if someone answered me or asked me about the situation, although despite everything I have sent I get no response. I just wanted someone to take me into account and that’s the normal thing to do but I’m sorry for that anyway.
I don’t want to bother the rest of the people or you guys just a reply or a show of interest, that’s all.
Unfortunately in 3 more hours I have to do a DEMO for one of the clients and the Avatar might not work properly so I don’t know what will happen from now on.
The ConveiChatBot component has an “OnFailure” event that gives me an error when the audio gives problems and with this I can ask again for the information that it did not give me. I’m still not completely sure that it works 100% of the times I get the error but at least it works better.
Azure voices cannot speak the characters ‘&’ and ‘<’.
This can be temporarily fixed by asking the character in your background to pronounce ‘&’ as ‘and’.
