Understanding n_batch in Llama.cpp for a Mobile AI Chat App

[Helldez]

Hi everyone,
I'm working on a mobile AI assistant app called d.ai (decentralized AI) that runs LLMs locally using Llama.cpp. My app processes chat requests in sequence, meaning there's no parallel request handling—just one user input at a time, generating a response before the next input is sent.

Given this setup, I'm wondering about the relevance of the n_batch parameter.

Does it have any impact in a scenario where requests are processed sequentially?
Is it essentially locked at 1 in such a case, making it an irrelevant parameter?
Or does it still affect things like token generation speed, memory usage, or CPU/GPU efficiency?

Any recommendations on best practices for managing batch size in this context (mobile)?

Thanks in advance!