| base_model: Qwen/Qwen3.5-2B | |
| library_name: peft | |
| tags: | |
| - lora | |
| - qwen | |
| - bash | |
| - shell | |
| - linux | |
| - text-generation | |
| # Qwen3.5-2B ShellCommand Linux LoRA | |
| This repository contains a PEFT LoRA adapter trained for Linux natural-language-to-shell translation. | |
| ## Artifact Type | |
| This is a LoRA adapter, not a merged full model checkpoint. | |
| ## Intended Behavior | |
| The model is tuned to return exactly one of: | |
| - a Bash command or short Bash snippet | |
| - `ASK: <one short clarifying question>` | |
| - `CANNOT: <brief reason>` | |
| ## Eval Snapshot | |
| - score: `276.5033` | |
| - verifier ok rate: `77.50%` | |
| - verifier command rate: `76.04%` | |
| - verifier ask rate: `75.00%` | |
| - verifier cannot rate: `100.00%` | |
| - exact any-exact rate: `25.00%` | |
| - exact parse-ok rate: `98.00%` | |
| ## Usage | |
| Load this adapter on top of `Qwen/Qwen3.5-2B` with PEFT. | |