AuricErgeson/shellwhiz-7b

ShellWhiz-7B: Natural Language to Shell Command Generation

ShellWhiz-7B is a specialized 7.6 billion parameter language model, fine-tuned from Qwen2.5-7B-Instruct, to translate natural language requests into shell commands. Developed by AuricErgeson, its primary purpose is to streamline command-line interactions by generating accurate find, grep, docker, git, and other utility commands from plain English descriptions.

Key Capabilities

• Comprehensive Command Coverage: Trained on 697 natural-language-to-shell-command pairs, it covers a wide range of operations including:
  • File and directory management (find, cp, mv, rm, chmod, du)
  • Text processing (grep, sed, awk, sort, cut)
  • Git workflows, Docker, and docker-compose commands
  • Process management (ps, kill, systemctl)
  • Networking utilities (curl, ssh, scp, ping)
  • Archiving and package management (tar, zip, apt, pip, npm)
• Placeholder Usage: The model intelligently uses placeholders like <directory> or <filename> for context-dependent values, rather than guessing specific paths.

Performance and Training

ShellWhiz-7B was trained using QLoRA (4-bit) on a single T4 GPU for 3 epochs, resulting in 0.53% trainable parameters (40.37 million). Evaluation on 105 held-out prompts showed 55.2% correct command generation and 95.2% syntax validity. Known limitations include occasional hallucinated flags, issues with negation, dropped constraints in multi-part requests, and sometimes retaining placeholders when concrete values are provided in the prompt. The training dataset, AuricErgeson/text-to-shell-dataset, was synthetically generated.