VAPI•2y ago

gpt-3.5-turbo & gpt-4 intermittently being called instead of gpt-4o

Our assistant is configured to use gpt-4o, but in call logs we see usage of other OpenAI models:

Model request started (gpt-3.5-turbo, openai)
Model request started (gpt-4-1106-preview-west-us, azure-openai, westus)

It seems like models we do not intend on using are being called. We see this issue whether we use our own API Key or the VAPI OpenAI key. If we use our own key, we can see gpt-3.5-turbo usage within our OpenAI usage dashboard.

Call IDs:
94261e17-d50d-4e87-bf44-666f62bcdf44 [ VAPI OpenAI Key]
b9fe4105-14dc-4792-b7d8-8f079a58e8de [Our own OpenAI Key]

Ask Inkeep•5/22/24, 2:18 PM

Hi @pakachan, I'm an AI assistant for Vapi. I'll analyze available sources and get back to you shortly.

Ask Inkeep•5/22/24, 2:18 PM

-----------------------------
While I generate an answer, here are some sources that might be helpful:

Web Calling
Update Assistant
Connecting Your Custom LLM to Vapi: A Comprehensive Guide
Outbound Calling
-----------------------------

Ask Inkeep•5/22/24, 2:18 PM

Answer

Sorry, I wasn't able to find any information about your question. You can get more help at Contact Us

AAsk Inkeep ### Answer Sorry, I wasn't able to find any information about your question. You...

Ask InkeepAPP•5/22/24, 2:18 PM

Got it. Tagging @Vapi Team for additional help.

living-lavender•5/22/24, 3:07 PM

Basically, we find the best model that has the least latency and perform intelligent model switching when you don't enter your key. If you don't want to, then simply add an empty fallbackModel parameter.

then-purple•5/22/24, 10:34 PM

Sahil, I set my fallback models to "[]" and set model as "gpt-4-0613". However, looking at the call log for call ID 03eb55c8-a1f3-4c4a-bac8-3c243e2fa85e, it is still using "gpt-4-1106-preview" and "gpt-3.5-turbo-1106". It doesn't look like "gpt-4-0613" was ever used looking through the log. Thanks.

then-purple•5/23/24, 2:06 AM

Here is another one, call ID 78c88842-0272-4ff0-8605-aa0ae4401370 for reference. I selected the gpt-4 model with no fallback models but got model requests for gpt-4o and gpt-3.5-turbo in some cases.

living-lavender•5/23/24, 2:08 AM

That's not possible if you set the fallback model as empty array and you have selected gpt-4o

living-lavender•5/23/24, 2:08 AM

you won't see any other model.

then-purple•5/23/24, 2:25 AM

Ah ok will try that thanks.

gpt-3.5-turbo & gpt-4 intermittently being called instead of gpt-4o

Answer

Similar Threads

Similar Threads

Answer

Similar Threads