[santiagobasulto]

But you'd need to keep both models in parallel, right? Using M1 to keep computing embeddings and using M2 for completions.