Okay, this is interesting! apparently I'm not the first one with the idea.

https://www.sciencedirect.com/science/article/pii/S016412122600049X

"Assessing small language models for code generation: An empirical study with benchmarks."

But SLMs are, in my opinion, still too big.

But I'm definitely curious about small footprint AIs for code assistance and automation of tedious computer tasks.