# Ethical Hacking LLM Fine-Tuning on Free Kaggle/Colab

## ✅ Updated Unsloth notebooks — root fixed

The notebooks now keep **Unsloth** for low VRAM, but fix the repeated Kaggle/Colab error:

```text
AttributeError: 'int' object has no attribute 'mean'
```

## What was fixed

That error came from a version mismatch between:

- `unsloth`
- `unsloth_zoo`
- `transformers`
- `trl`

plus Kaggle reusing old compiled code from:

```text
/kaggle/working/unsloth_compiled_cache
```

Each notebook now:

1. Deletes the stale Unsloth compiled cache.
2. Uninstalls and reinstalls latest `unsloth` + `unsloth_zoo` together.
3. Forces a one-time kernel restart after install.
4. Uses modern TRL `SFTConfig` API.
5. Uses `processing_class=tokenizer` instead of older `tokenizer=` trainer argument.
6. Sets `trainer.model_accepts_loss_kwargs = False` to prevent `num_items_in_batch` leakage.

## Notebooks

| Notebook | Model | Best use |
|---|---|---|
| [`EthicalHacking_LFM2.5_Ultimate_Colab.ipynb`](https://huggingface.co/asdf98/ethical-hacking-llm-colab/blob/main/EthicalHacking_LFM2.5_Ultimate_Colab.ipynb) | `unsloth/LFM2.5-1.2B-Instruct` | **Recommended first**, fastest and lowest VRAM |
| [`EthicalHacking_Qwen3-4B_Ultimate_Colab.ipynb`](https://huggingface.co/asdf98/ethical-hacking-llm-colab/blob/main/EthicalHacking_Qwen3-4B_Ultimate_Colab.ipynb) | `unsloth/Qwen3-4B-Instruct-2507-unsloth-bnb-4bit` | Stronger model, tighter T4 VRAM |
| [`EthicalHacking_Gemma4_E2B_Colab.ipynb`](https://huggingface.co/asdf98/ethical-hacking-llm-colab/blob/main/EthicalHacking_Gemma4_E2B_Colab.ipynb) | `unsloth/gemma-4-E2B-it-unsloth-bnb-4bit` | Experimental; use LFM2.5 first |
| [`EthicalHacking_Stable_QLoRA_ManualLoop_NO_UNSLOTH.ipynb`](https://huggingface.co/asdf98/ethical-hacking-llm-colab/blob/main/EthicalHacking_Stable_QLoRA_ManualLoop_NO_UNSLOTH.ipynb) | Vanilla PEFT QLoRA | Backup if Unsloth breaks again |

## Important run instructions

In Kaggle/Colab:

1. Upload/open the notebook.
2. Run cell 1.
3. Cell 1 will install/update Unsloth and intentionally restart the kernel.
4. After restart, run all cells from the top again.
5. Do **not** skip cell 1. It clears the compiled cache that caused the bug.

## Fast smoke test

Before full training, set:

```python
SAMPLE_SIZE = 1000
MAX_STEPS = 10
```

If that trains, increase to the defaults.

## Safety

Use only for ethical, defensive, authorized cybersecurity education and research.