AryaYT/nl2shell-0.8b is a 0.8B-parameter model fine-tuned from Qwen3.5-0.8B with QLoRA on the NL2Bash benchmark (~10,000 pairs) plus 40 hand-crafted macOS command pairs. It outputs only the shell command — no markdown, no explanation.

Running on CPU (free HF Spaces tier). Generation takes 5-20 seconds. For instant results, duplicate this Space and choose a GPU tier, or run the model locally via Ollama.

0 1
16 256

Example Prompts

Click an example to run it
Pages:
Property Value
Base model Qwen/Qwen3.5-0.8B
Fine-tuning QLoRA (4-bit, rank 16, alpha 32)
Training data NL2Bash + 40 macOS synthetic pairs
Format ChatML
Max output 128 tokens
GGUF (edge) q4_k_m ~400 MB, q8_0 ~650 MB
License MIT

Run locally with Ollama:

ollama run hf.co/AryaYT/nl2shell-0.8b

Note: Always review shell commands before running them, especially those involving rm, kill, sudo, or network operations. This model is a research demo — treat output as a starting point, not ground truth.