Default Stop Sequence is not working when user provides additional stop sequences

[Nanthagopal-Eswaran]

What is the issue?

I want to stop generation at stop sequence "\nObservation".

My expectation here is generation will stop at either "\nObservation" or models default stop sequence "<|eot_id|>" (in llama3 case).

But it does not happen and model keep on generating answer.

Request:

Response:

I think we should add stop sequences to default stop sequence instead of replacing them

OS

Windows

GPU

No response

CPU

Intel

Ollama version

0.1.27