Understanding the need to serve efficiently, I was wondering if anybody here has experience with merging the mistral model trained with unsloth using PEFT back into the original mistral ai model for compatibility and serving via vllm framework?

Best Regards, Lukas