{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d5f29e43-facd-4996-ae82-f9b5f271fd3a",
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install -q unsloth transformers datasets trl huggingface_hub"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6f2c9817-d609-44b5-b68c-e8d129cbf51d",
   "metadata": {},
   "source": [
    "#Installation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "c305abc3-fcd5-4000-86b1-c10adebbb197",
   "metadata": {},
   "outputs": [],
   "source": [
    "from unsloth import FastLanguageModel, is_bfloat16_supported\n",
    "import os, json\n",
    "from datasets import load_dataset\n",
    "from trl import SFTTrainer\n",
    "from transformers import TrainingArguments"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e5d28b79-80e8-4444-896e-1fd91af6b0eb",
   "metadata": {},
   "source": [
    "#Unsloth"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f6b96d82-d19d-4da7-a2c1-f48ff01085c9",
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_FILE = \"Anthro_ontologies.json\"\n",
    "\n",
    "MODEL_NAME = \"unsloth/gemma-2-2b\"\n",
    "OUTPUT_DIR = \"outputs\"\n",
    "\n",
    "MAX_SEQ_LEN = 2048\n",
    "BATCH_SIZE = 2\n",
    "GRAD_ACCUM = 4\n",
    "LR = 2e-4"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aaf7f082-de66-4371-bcbb-2e7dcb097680",
   "metadata": {},
   "source": [
    "#Data Prep"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9d3fbf8f-ae8e-455d-b9ab-c74c07b36ae3",
   "metadata": {},
   "outputs": [],
   "source": [
    "model, tokenizer = FastLanguageModel.from_pretrained(\n",
    "    model_name=MODEL_NAME,\n",
    "    max_seq_length=MAX_SEQ_LEN,\n",
    "    load_in_4bit=True,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f983d204-64ce-46e2-a8e6-f1962518f815",
   "metadata": {},
   "source": [
    "#Train the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ab6a3db2-400b-48ff-b4dd-52d55bb286f9",
   "metadata": {},
   "outputs": [],
   "source": [
    "EOS_TOKEN = tokenizer.eos_token or \"</s>\"\n",
    "\n",
    "def format_prompt(x):\n",
    "    return {\n",
    "        \"text\": f\"\"\"Write a response that appropriately completes the request.\n",
    "\n",
    "### Instruction:\n",
    "{x[\"instruction\"]}\n",
    "\n",
    "### Input:\n",
    "{x[\"input\"]}\n",
    "\n",
    "### Response:\n",
    "{x[\"output\"]}\"\"\" + EOS_TOKEN\n",
    "    }\n",
    "\n",
    "ds = load_dataset(\"json\", data_files=DATA_FILE, split=\"train\")\n",
    "ds = ds.train_test_split(test_size=0.02, seed=3407)\n",
    "ds_train, ds_val = ds[\"train\"], ds[\"test\"]\n",
    "\n",
    "ds_train = ds_train.map(format_prompt)\n",
    "ds_val   = ds_val.map(format_prompt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "0b7647b2-1512-4969-9ff9-71f9c8281903",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = FastLanguageModel.get_peft_model(\n",
    "    model,\n",
    "    r=16,\n",
    "    lora_alpha=16,\n",
    "    lora_dropout=0,\n",
    "    bias=\"none\",\n",
    "    use_gradient_checkpointing=\"unsloth\",\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "466271b0-9c44-4b19-8f8c-1398ed9a6d46",
   "metadata": {},
   "outputs": [],
   "source": [
    "trainer = SFTTrainer(\n",
    "    model=model,\n",
    "    tokenizer=tokenizer,\n",
    "    train_dataset=ds_train,\n",
    "    eval_dataset=ds_val,\n",
    "    dataset_text_field=\"text\",\n",
    "    max_seq_length=MAX_SEQ_LEN,\n",
    "    args=TrainingArguments(\n",
    "        output_dir=OUTPUT_DIR,\n",
    "        per_device_train_batch_size=BATCH_SIZE,\n",
    "        gradient_accumulation_steps=GRAD_ACCUM,\n",
    "        learning_rate=LR,\n",
    "        num_train_epochs=3,\n",
    "        fp16=not is_bfloat16_supported(),\n",
    "        bf16=is_bfloat16_supported(),\n",
    "        logging_steps=5,\n",
    "    ),\n",
    ")\n",
    "\n",
    "trainer.train()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "678c8480-6f59-4a73-84bb-872709140d7d",
   "metadata": {},
   "source": [
    "#Save + Upload to HF"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "9bbaa447-4a03-45a1-b623-f925b320473e",
   "metadata": {},
   "outputs": [],
   "source": [
    "from huggingface_hub import login, HfApi\n",
    "\n",
    "login()  # paste HF token\n",
    "\n",
    "REPO_ID = \"your-username/your-model-name\"\n",
    "\n",
    "trainer.model.save_pretrained(OUTPUT_DIR)\n",
    "tokenizer.save_pretrained(OUTPUT_DIR)\n",
    "\n",
    "api = HfApi()\n",
    "api.create_repo(repo_id=REPO_ID, repo_type=\"model\", exist_ok=True)\n",
    "\n",
    "api.upload_folder(\n",
    "    folder_path=OUTPUT_DIR,\n",
    "    repo_id=REPO_ID,\n",
    "    repo_type=\"model\",\n",
    ")\n",
    "\n",
    "print(\"Upload complete\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.12.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}