Trained a 1.5B to write shell commands so I'd stop googling tar flags. Runs on a laptop CPU in ~1 sec.
Heat trend
The percentage is based on available heat signal, not comment count or independent people.
This covers a coding tool or code-capability update — useful for developers assessing workflow changes and reusable value.
I've been googling "tar extract gz" for about ten years. and I finally did something about it.
It started out as a research project and I ended up with a Fine-tuned Qwen2.5-Coder-1.5B on 125k natural-language/command pairs, merged and quantized to Q4_K_M. 941MB which runs through llama.cpp. On my laptop (i5-11320H, 4 threads): 31.9 tok/s, 0.59s median per query, 1.6GB RAM.
I benchmarked it and it scores 0.620 on InterCode-ALFA. Untuned Qwen2.5-Coder-7B gets 0.613, GPT-4o gets 0.73. Not frontier, but it's roughly a 7B's answer at a quarter the parameters on a CPU. Theres a 3B variant too that scores higher.
There's also few static safety checker, because it will absolutely write a command that wipes your root if you ask it to:
I have published the weights: huggingface.co/ThorOdinson246/nl2sh-1.5b-Q4_K_M and Code: github.com/ThorOdinson246/whatisit-nl2sh. I posted few days ago in LocalLLM and it did well 300+ stars and so many good suggestions so I figured people here will be interested too.
Both Apache-2.0. If you want to poke holes in the method or you've got ideas, please comment or open a PR. A ⭐ helps if you find it useful.