Shell Buddy (GGUF)

A small model that turns plain English into a single Linux bash command. Fine-tuned from Qwen2.5-0.5B-Instruct and quantized to 8-bit (531 MB), so it runs on an ordinary laptop CPU.

Try it

ollama run hf.co/SushmithaJujare/shell-buddy-GGUF "find all files bigger than 1GB in my home folder"

Results (100 unseen examples)

Model Right tool chosen Exact match
Base Qwen 0.5B (not fine-tuned) 62% 5%
Qwen 7B (14ร— bigger, not fine-tuned) 63% 9%
Shell Buddy (0.5B, fine-tuned, 8-bit) 91% 18%

How it was made

  • ~10,800 English โ†’ bash pairs cleaned from NL2Bash
  • LoRA fine-tuning on a free Google Colab GPU
  • Merged and converted to GGUF (q8_0) with llama.cpp

Limitations

  • It can be wrong. Always read a command before running it.
  • Weaker on everyday commands like pwd, df -h and du, because most of the training data uses find.
Downloads last month
72
GGUF
Model size
0.5B params
Architecture
qwen2
Hardware compatibility
Log In to add your hardware

8-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for SushmithaJujare/shell-buddy-GGUF

Quantized
(395)
this model

Dataset used to train SushmithaJujare/shell-buddy-GGUF