---
license: gemma
base_model:
- google/gemma-3-1b-it
tags:
- gemma
- terminal
- linux
- command-generation
- sft
- Mica
- fine-tuned
pipeline_tag: text-generation
library_name: transformers
---

# Gemma 3 1B Terminal Assistant

A fine-tuned version of Google's **Gemma 3 1B Instruction Tuned model** specialized for terminal command generation.

This model was trained to understand natural language requests and generate safe, minimal terminal commands.

## Model Details

Base Model:

- google/gemma-3-1b-it

Fine-tuning Method:

- Supervised Fine-Tuning (SFT)
- Mica fine-tuning
- Mica merged into the base model

Training Dataset:

- mshojaei77/terminal-command-execution-sft

Task:

- Linux terminal commands
- Windows commands
- Shell scripting
- Command explanation
- Safe command generation


## Example Usage


### Python Example

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch


model_name = "Paulwalker4884/gemma-3-1b-terminal-assistant"


tokenizer = AutoTokenizer.from_pretrained(model_name)

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto"
)


messages = [
    {
        "role": "system",
        "content": "You are a safe terminal command assistant."
    },
    {
        "role": "user",
        "content": "Find all python files recursively"
    }
]


inputs = tokenizer.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)


outputs = model.generate(
    inputs,
    max_new_tokens=128,
    temperature=0.2
)


response = tokenizer.decode(
    outputs[0][inputs.shape[-1]:],
    skip_special_tokens=True
)


print(response)
```

---

## Example

Input:

```
Find all python files recursively
```

Output:

```bash
find . -name "*.py"
```


Input:

```
Find all log files modified in the last 7 days and save them
```

Output:

```bash
find . -name "*.log" -mtime -7 > recent_logs.txt
```


## Safety

This model is trained to avoid blindly generating destructive commands.

For potentially dangerous operations, users should verify commands before execution.


## Limitations

- The model may generate incorrect commands.
- Always review generated commands before running them.
- Performance depends on the quality of the input prompt.


## Training Information

Dataset size:

- Train: 31,429 examples
- Evaluation: 239 examples





Frameworks:

- Transformers
- TRL
- PEFT
- PyTorch

BY : Yasin Keykha