This character 2786b252-9d59-11f0-88d8-42010a7be025 has from possibly last night but definitely this morning has suddenly started to have much more failed responses , it’s about every 1 in three response right now so pretty much usuable, as the video of a Playground chat from 10 minutes ago shows (it’s the same in my app).
The character is about 2 weeks old and was working fine when there was a low amount of people with access to the character, then the character was made available to all users of my app yesterday which seems to roughly coincide with when the increased failed response rate has started happening, unsure if it’s connected, tbh the character was very reliable before, more so than usual and with other characters.
When the failed responses happen I’ve noticed my voice is still being picked up because I am seeing it being transcribed in a chat bubble, and when the failed responses happen my app is still receiving emotion data from the servers.
I am aware this could be an LLM issue so out of Convai’s control. Not sure how to check for Llama issues, I know there’s a website to check if ChatGPT is operating normally.
We’re sorry about the issue you’re experiencing. We’re currently investigating it, and we’ll update you here as soon as it’s resolved. Thank you for your patience.
I know you’re doing your best and working on it @K3 just wanted to update you that this is affecting all characters now it seems (it might have been all along), not just the character ID I gave above. I’ve noticed sometimes the emotion data is not received too.
This appears to be an issue related to the model provider. We’ll investigate it in detail and provide a fix. In the meantime, please try using non-NSFW models.
I tested with this character 2a30e30c-69ce-11ef-80bd-42010a7be011 and here’s the Session ID 2507abc151e837096c696fd5c607dfd0
I found they were still censored, not completely but mostly. The non-responses were still happening in about 1 in every 5 character replies, emotion data was being received ok.
Hi Tyke,
We have a few questions. Since you seem to be using Convai in a production setting, can you clarify a few points for us? These will help us debug the non-responses issue.
What is the concurrency that you are expecting on your app, i.e, how many customers will be conversing with your character simultaneously?
What is the average number of requests being hit by your app to our Convai servers?
Also, a few tips and suggestions regarding the Censorship:
Asking directly to the character whether they are censored or not, is not a very good method of testing for censorship.
Instead try having a simple conversation which is directed towards uncensored dialogue and contents. This will be a more real world test than asking it speak, or say curse words.
Try using stronger or more aggressive prompts in the speaking style and the sample dialogues that you have set if the character refuses to behave in an uncensored manner.
There are about 1050 users of my app (they have valid login details), I am not sure how many use it at the same time sorry.
The interactions reached for the last month is about 160,000
Thank you for the tips, I have taken note. It seems when uncensored Llama is the selected LLM, the avatar often doesn’t respond, although emotion data is received.
