https://vapi.ai/ logo
Join Discord
Powered by
Assistant ends call in the middle of the conversat...
# support
l
Not sure how to debug this but seems like the conversation ends in the middle by the assistant. Assistant ends call when the conversation is like: Assistant: What language would you like to see the articles in? User: French Assitant: Good bye Here are call ids: 14589478-9c36-4247-8c17-c8a3447fddc5 08702490-6dca-4eee-9c4a-7d0929ad9d1b
v
Hi @Lacoste, I'm an AI assistant for Vapi. I'll analyze available sources and get back to you shortly.
----------------------------- **While I generate an answer, here are some sources that might be helpful**: * [Delete Assistant]() * [List Assistants]() * [Create Assistant]() * [Create Phone Call]() -----------------------------
### Answer It seems like your assistant is ending calls prematurely due to an "end-call phrase" being triggered. Even if the phrase is not explicitly said by the assistant in the conversation, it might be part of the system prompt or another embedded function causing this behavior. To debug this, you can review the assistant's configuration, especially the "endCallPhrases" and "system prompt" settings. Ensure that the phrases intended to end calls are correctly configured and not inadvertently used in a way that might trigger unintended call terminations. If you've included phrases in your system prompt or conversation that might resemble or match your configured "end-call phrases," this could be why calls are ending unexpectedly. Consider modifying these phrases or the conditions under which they trigger the call to end. For further insights into Vapi's handling of calls and the configuration of assistants that could potentially cause unexpected call endings, reviewing the official documentation on creating, updating, and managing assistants could prove helpful [(1)]() , [(2)]() .
m
what model are you using?
l
FYI, I do have endCallPhrases as "Have a great day"
m
I hear this a lot with 3.5 x models
l
Oh, I was using Groq, llama 3 70b
I'm curious if vAPI has special end-call phrases other than what I provided that is triggering this issue.
m
Nah it's just misfiring the function
3.5 and equivalent models or under without fine tuning kinda suck at functions. But also might be a error on vapi side I'm sure they'll check soon
s
Use GPT-4 cause llama doesn't support function calling via streaming.
l
Hmmm, I didn't set any function calling for the model.
m
The end call function is a function
Just predefined as opposed to you coding it
l
Ah I see, it gets added to the model
m
Yessir
l
Thanks for the support! I guess I can remove the endCallFunction or change the model.
m
Yeah you can try 3.5 0125 and see how it performs if you're trying to keep cost down
What's your use case btw what are you building
l
I'm building for a client to assist users onboarding / setup who are on a free trial as there is no support for them currently
m
Nice, do you think you or your client would ever need the ability to facilitate or process / manage payments?
l
Not sure atm. I'm just a developer building stuff for the company. I dont' talk direclty to client
s
hey mason, would be great for our project if you could share training files. We are having same issue
m
Depends on your use case but I can definetly send you some examples if you'd like boss
s
yes please Mason that would be great bud
m
What functionality are you trying to accomplish with your functions?
s
we have enabled the endCallFunction and it keeps randomly ending the call abruptly. No issue when we disable the function. Need a work around or to understand better what we are doing wrong
m
Yeah fine tuning is more for custom functions, for that default function your only real choice is using gpt 4
p
Hey @Mason | Building KOI Would you be down to share that training file? I am interesting in fine tuning my 3.5-turbo setup too...
m
Of course man, what’s your use case
p
We're doing orders over the phone.
m
Gotcha so fine tuning models in general or fine tuning for functions?
p
Just functions really, so we can avoid having GPT-4's higher cost and use some custom functions. Looks like that would not help us with VAPI functions like end-call/forward-call tho right?
a
I'd love to see those training files too pls, would like to see the tuning in general and functions!
l
Would be great if you could send me the file too! I am working on scam prevention calls agent
m
Sounds good! I'm going to open source some datasets on my github this coming week I'll tag you guys when I launch it
@Lukas Narus
Correct yeah it would only be custom functions
5 Views
PreviousNext
Powered by