Above is a screenshot of the Flowise build we are currently trying to PoC, and we are struggling with a model to replace OpenAI 3.5 to help us do the calling of the API.
We are essentially trying to build a much cheaper version of GPTs with its function calling and RAG abilities, as we have a huge amount of traffic that will be coming through our solution.
However, the main issue we have is replacing the language model to be able to make the API call to our Webhook. OpenAI’s GPT 3.5 seems to work fairly reliably, but we have tried swapping this out with:
- Mistral 7b
- Mistral Instruct 7b
- Zephyr 7b beta
- Llama 7b
But we have been unable to get the AI to call the webhook.
Looking for recommendations on how we can deal with this using smaller Open-Source LLM models!
Your help is appreciated