gemma.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Loading Lora Configuration"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from peft import LoraConfig\n",
    "\n",
    "lora_config = LoraConfig(\n",
    "    r=8,\n",
    "    target_modules=[\"q_proj\", \"o_proj\", \"k_proj\", \"v_proj\", \"gate_proj\", \"up_proj\", \"down_proj\"],\n",
    "    task_type=\"CAUSAL_LM\",\n",
    ")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Loading Model and Tokenizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ubuntu/Desktop/GEMMA_FINETUNE/.venv/lib/python3.11/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
      "  from .autonotebook import tqdm as notebook_tqdm\n",
      "`config.hidden_act` is ignored, you should use `config.hidden_activation` instead.\n",
      "Gemma's activation function will be set to `gelu_pytorch_tanh`. Please, use\n",
      "`config.hidden_activation` if you want to override this behaviour.\n",
      "See https://github.com/huggingface/transformers/pull/29402 for more details.\n",
      "Loading checkpoint shards: 100%|██████████| 2/2 [00:01<00:00,  1.03it/s]\n"
     ]
    }
   ],
   "source": [
    "import torch\n",
    "from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig\n",
    "\n",
    "model_id = \"google/gemma-2b\"\n",
    "bnb_config = BitsAndBytesConfig(\n",
    "    load_in_4bit=True,\n",
    "    bnb_4bit_quant_type=\"nf4\",\n",
    "    bnb_4bit_compute_dtype=torch.bfloat16\n",
    ")\n",
    "\n",
    "hf_token = \"ADD_TOKEN_HERE\" # also take permission from gemma from huggingface\n",
    "\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_id, token=hf_token)\n",
    "model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config, device_map={\"\":0}, token=hf_token)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Loading Datasets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import load_dataset\n",
    "\n",
    "data = load_dataset(\"Yemmy1000/cybersec_embedding_llama_chat\")\n",
    "data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Setting Training Parameters and Fine-Tuning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import transformers\n",
    "from trl import SFTTrainer\n",
    "\n",
    "def formatting_func(examples):\n",
    "    texts = []\n",
    "    instructions = examples['INSTRUCTION']\n",
    "    responses = examples['RESPONSE']\n",
    "\n",
    "    for idx in range(len(instructions)):\n",
    "        instruction = instructions[idx].split('\\n')[1]\n",
    "        response = responses[idx]\n",
    "        text = f\"Instruction: {instruction} \\n Output: {response}<eos>\"\n",
    "        texts.append(text)\n",
    "    \n",
    "    return texts\n",
    "\n",
    "trainer = SFTTrainer(\n",
    "    model=model,\n",
    "    train_dataset=data[\"train\"],\n",
    "    args=transformers.TrainingArguments(\n",
    "        per_device_train_batch_size=1,\n",
    "        gradient_accumulation_steps=4,\n",
    "        warmup_steps=100,\n",
    "        num_train_epochs=4,\n",
    "        learning_rate=2e-4,\n",
    "        fp16=False,\n",
    "        logging_steps=200,\n",
    "        output_dir=\"outputs\",\n",
    "        optim=\"paged_adamw_8bit\"\n",
    "    ),\n",
    "    peft_config=lora_config,\n",
    "    formatting_func=formatting_func,\n",
    ")\n",
    "\n",
    "trainer.train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "trainer.save_model('./Model')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "7e0c02a2059a451ebd0c8b8726b68003",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "Loading checkpoint shards:   0%|          | 0/2 [00:00<?, ?it/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import torch\n",
    "from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig\n",
    "\n",
    "model_id = \"./final\"\n",
    "bnb_config = BitsAndBytesConfig(\n",
    "    load_in_4bit=True,\n",
    "    bnb_4bit_quant_type=\"nf4\",\n",
    "    bnb_4bit_compute_dtype=torch.bfloat16\n",
    ")\n",
    "\n",
    "\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_id)\n",
    "model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config, device_map={\"\":0})"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}