CyberRanger/colab_notebooks/RangerBot_V5_Colab_Trainer-V2.ipynb

{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "header"
      },
      "source": [
        "# 🎖️ RangerBot V5: The QLoRA Forge\n",
        "\n",
        "**NCI H9AIMLC - AI/ML in Cybersecurity**\n",
        "**Commander:** David Keane (x20188559)\n",
        "\n",
        "Fine-tune RangerBot V5 into Llama, Qwen, or SmolLM."
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "installs"
      },
      "outputs": [],
      "source": [
        "# 1. Install Unsloth\n",
        "%%capture\n",
        "!pip install \"unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git\"\n",
        "!pip install --no-deps \"xformers<0.0.27\" \"trl<0.9.0\" peft accelerate bitsandbytes\n",
        "print(\"✅ Installed\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "config"
      },
      "outputs": [],
      "source": [
        "# 2. Configuration\n",
        "from unsloth import FastLanguageModel\n",
        "import torch\n",
        "\n",
        "# CHOOSE ONE:\n",
        "MODEL_NAME = \"unsloth/SmolLM2-1.7B-Instruct-bnb-4bit\"\n",
        "\n",
        "model, tokenizer = FastLanguageModel.from_pretrained(\n",
        "    model_name = MODEL_NAME,\n",
        "    max_seq_length = 2048,\n",
        "    load_in_4bit = True,\n",
        ")\n",
        "print(f\"✅ Loaded {MODEL_NAME}\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "lora"
      },
      "outputs": [],
      "source": [
        "# 3. Add LoRA\n",
        "model = FastLanguageModel.get_peft_model(\n",
        "    model,\n",
        "    r = 16,\n",
        "    target_modules = [\"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\", \"gate_proj\", \"up_proj\", \"down_proj\"],\n",
        "    lora_alpha = 16,\n",
        "    lora_dropout = 0,\n",
        "    bias = \"none\",\n",
        ")\n",
        "print(\"✅ LoRA Ready\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "data"
      },
      "outputs": [],
      "source": [
        "# 4. Load Data\n",
        "from datasets import Dataset\n",
        "import json\n",
        "\n",
        "# UPLOAD training_data_cyberranger.json to Colab first!\n",
        "with open('training_data_cyberranger.json', 'r') as f:\n",
        "    raw_data = json.load(f)\n",
        "\n",
        "def format_prompts(examples):\n",
        "    texts = [f\"### Instruction:\\n{i}\\n\\n### Response:\\n{o}\" for i, o in zip(examples[\"instruction\"], examples[\"output\"])]\n",
        "    return { \"text\" : texts }\n",
        "\n",
        "dataset = Dataset.from_list(raw_data[\"training_data\"])\n",
        "dataset = dataset.map(format_prompts, batched = True)\n",
        "print(\"✅ Data Loaded\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "train"
      },
      "outputs": [],
      "source": [
        "# 5. Train\n",
        "from trl import SFTTrainer\n",
        "from transformers import TrainingArguments\n",
        "from unsloth import is_bfloat16_supported\n",
        "\n",
        "trainer = SFTTrainer(\n",
        "    model = model,\n",
        "    tokenizer = tokenizer,\n",
        "    train_dataset = dataset,\n",
        "    dataset_text_field = \"text\",\n",
        "    max_seq_length = 2048,\n",
        "    args = TrainingArguments(\n",
        "        per_device_train_batch_size = 2,\n",
        "        gradient_accumulation_steps = 4,\n",
        "        max_steps = 120, # 🎖️ HEAT INCREASED TO 120 STEPS\n",
        "        learning_rate = 2e-4,\n",
        "        fp16 = not is_bfloat16_supported(),\n",
        "        bf16 = is_bfloat16_supported(),\n",
        "        logging_steps = 1,\n",
        "        output_dir = \"outputs\",\n",
        "    ),\n",
        ")\n",
        "trainer.train()\n",
        "print(\"✅ Training Complete\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "stabilized_chat"
      },
      "outputs": [],
      "source": [
        "# 6. Direct Comm-Link (Stabilized)\n",
        "from unsloth import FastLanguageModel\n",
        "FastLanguageModel.for_inference(model)\n",
        "\n",
        "print(\"🎖️ CyberRanger V5-GENESIS (Stabilized) Online. Type exit to quit.\")\n",
        "while True:\n",
        "    user_input = input(\"Commander: \")\n",
        "    if user_input.lower() in [\"exit\", \"quit\", \"bye\"]: break\n",
        "\n",
        "    prompt = f\"### Instruction:\\n{user_input}\\n\\n### Response:\\n\"\n",
        "    inputs = tokenizer([prompt], return_tensors = \"pt\").to(\"cuda\")\n",
        "    outputs = model.generate(\n",
        "        **inputs, \n",
        "        max_new_tokens = 128, \n",
        "        temperature = 0.7, \n",
        "        repetition_penalty = 1.2, \n",
        "        do_sample = True,\n",
        "        pad_token_id = tokenizer.eos_token_id\n",
        "    )\n",
        "    \n",
        "    response = tokenizer.batch_decode(outputs)[0]\n",
        "    final_response = response.split(\"### Response:\\n\")[-1].replace(tokenizer.eos_token, \"\").strip()\n",
        "    print(f\"\\nCyberRanger: {final_response}\\n\" + \"-\"*60 + \"\\n\")"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "id": "gguf_export"
      },
      "outputs": [],
      "source": [
        "# 7. Save to GGUF\n",
        "model.save_pretrained_gguf(\"model\", tokenizer, quantization_method = \"q4_k_m\")\n",
        "print(\"✅ GGUF Saved in model/ folder\")"
      ]
    }
  ],
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "gpuType": "T4"
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}