from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load model (base + adapters)
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3-0.6B")
model = PeftModel.from_pretrained(base_model, "Eng-Elias/qwen3-0.6b-terminal-instruct")
tokenizer = AutoTokenizer.from_pretrained("Eng-Elias/qwen3-0.6b-terminal-instruct")

# Generate command
def generate_command(instruction, os_tag="[LINUX]"):
    prompt = f"### Instruction:\n{instruction}\n\n### Input:\n{os_tag}\n\n### Response:\n"
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(**inputs, max_new_tokens=100, do_sample=False)
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response.split("### Response:")[-1].strip()

# Examples
print(generate_command("List all files including hidden ones", "[LINUX]"))
# Output: ls -a

print(generate_command("Create a new folder named projects", "[WINDOWS]"))
# Output: mkdir projects

print(generate_command("Show disk usage", "[MAC]"))
# Output: df -h

Features

OS-Specific Commands

Generate commands tailored for specific operating systems:

python

# Linux
generate_command("Find all Python files", "[LINUX]")
# find . -name '*.py'

# Windows  
generate_command("Find all Python files", "[WINDOWS]")
# dir /s /b *.py

# macOS
generate_command("Find all Python files", "[MAC]")
# find . -name '*.py'

JSON Output (All OS)

Get commands for all operating systems in JSON format:

generate_command("Delete file named temp.txt", "Return the command for all operating systems as JSON")

Output:

json

{
  "description": "Delete file named temp.txt",
  "linux": "rm temp.txt",
  "windows": "del temp.txt",
  "mac": "rm temp.txt"
}

Supported Command Categories

Category	Examples
File Operations	list, copy, move, delete, find, rename
Directory Operations	create, remove, navigate, list contents
System Info	disk usage, memory, CPU, processes
Text Processing	grep, sed, awk, sort, uniq
Network	ping, curl, wget, netstat, ssh
Compression	tar, zip, gzip, unzip
Permissions	chmod, chown, icacls
Package Management	apt, yum, brew, choco

Prompt Format

The model uses an Alpaca-style prompt format:

### Instruction:
{natural language description of the command}

### Input:
{OS tag or special request}

### Response:
{generated terminal command}

Input Options

Input	Description
`[LINUX]`	Generate Linux/Unix command
`[WINDOWS]`	Generate Windows command
`[MAC]`	Generate macOS command
`Return the command for all operating systems as JSON`	Get JSON with all OS commands

Training Details

Parameter	Value
Dataset	Custom terminal command dataset with 10000+ examples
Base Model	Qwen/Qwen3-0.6B
Method	QLoRA (4-bit NF4)
LoRA Rank	16
LoRA Alpha	32
Target Modules	q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Learning Rate	2e-4
Training Steps	~1800
Batch Size	4 (×4 gradient accumulation)

Related Models

LoRA Adapters: Eng-Elias/qwen3-0.6b-terminal-instruct-lora

Experiment Tracking

Training metrics and hyperparameters are logged to Weights & Biases:

Project: qwen3-terminal-instruct
Run: qwen3-0.6b-terminal-20251230_2244
Dashboard: View on W&B

Limitations

Commands are based on common usage patterns; complex or obscure commands may not be accurate
The model may occasionally generate slightly different but functionally equivalent commands
JSON output format is consistent but may vary in structure for edge cases
Trained primarily on English instructions

License

This project is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).

You are free to:

Share — copy and redistribute the material in any medium or format
Adapt — remix, transform, and build upon the material

Under the following terms:

Attribution — You must give appropriate credit
NonCommercial — You may not use the material for commercial purposes
ShareAlike — If you remix, transform, or build upon the material, you must distribute your contributions under the same license

Full license: https://creativecommons.org/licenses/by-nc-sa-4.0/legalcode

Citation

bibtex

@misc{qwen3-terminal-instruct,
  author = {Eng-Elias},
  title = {Qwen3-0.6B Terminal Command Generator},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/Eng-Elias/qwen3-0.6b-terminal-instruct}
}

Acknowledgments

Qwen Team for the base model
Hugging Face for the transformers and PEFT libraries