{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !pip install nest_asyncio \\\n",
    "#     langchain_openai langchain_huggingface langchain_core langchain langchain_community langchain-text-splitters \\\n",
    "#     python-pptx==1.0.2 nltk==3.9.1 pymupdf lxml \\\n",
    "#     sentence-transformers IProgress \\\n",
    "#     huggingface_hub ipywidgets \\\n",
    "#     qdrant-client langchain_experimental\n",
    "\n",
    "# !pip install sentence_transformers datasets pyarrow\n",
    "# !pip install torch\n",
    "# !pip install accelerate>=0.26.0\n",
    "# !pip install transformers\n",
    "# !pip install wandb\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "import nest_asyncio\n",
    "\n",
    "nest_asyncio.apply()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "#!pip install -qU langchain_openai langchain_huggingface langchain_core langchain langchain_community langchain-text-splitters"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "#!pip install -qU faiss-cpu python-pptx==1.0.2 nltk==3.9.1 pymupdf beautifulsoup4 lxml"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "#!pip install -qU sentence-transformers\n",
    "#!pip install -qU IProgress\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import getpass\n",
    "\n",
    "os.environ[\"OPENAI_API_KEY\"] = getpass.getpass(\"Enter Your OpenAI API Key: \")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "hf_username = getpass.getpass(\"Enter Your Hugging Face Username: \")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "df7fbe16b4c44797abc886b87583af59",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "VBox(children=(HTML(value='<center> <img\\nsrc=https://huggingface.co/front/assets/huggingface_logo-noborder.sv…"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from huggingface_hub import notebook_login\n",
    "notebook_login()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'type': 'user', 'id': '67624d1b57e77fe6e0c87ae5', 'name': 'drewgenai', 'fullname': 'Drew DeMarco', 'email': 'drewgenai@gmail.com', 'emailVerified': True, 'canPay': False, 'periodEnd': None, 'isPro': False, 'avatarUrl': 'https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/L6eLaZmCK4jqW3ZTLYIAR.png', 'orgs': [], 'auth': {'type': 'access_token', 'accessToken': {'displayName': 'newotken', 'role': 'write', 'createdAt': '2025-02-12T04:11:04.130Z'}}}\n"
     ]
    }
   ],
   "source": [
    "from huggingface_hub import whoami\n",
    "print(whoami())\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "mkdir: cannot create directory ‘example_files’: File exists\n",
      "mkdir: cannot create directory ‘output’: File exists\n"
     ]
    }
   ],
   "source": [
    "!mkdir example_files\n",
    "!mkdir output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain_community.document_loaders import DirectoryLoader\n",
    "from langchain_community.document_loaders import PyMuPDFLoader\n",
    "\n",
    "path = \"example_files/\"\n",
    "text_loader = DirectoryLoader(path, glob=\"*.pdf\", loader_cls=PyMuPDFLoader)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "1️⃣ Header-Based Chunking (Title-Based Splitter)\n",
    "Uses document structure to split on headings, section titles, or patterns.\n",
    "Works well for structured documents with named assessments, numbered lists, or headers.\n",
    "Example: If it detects Chronic Pain Adjustment Index (CPAI-10), it groups everything under that title.\n",
    "2️⃣ Semantic Chunking (Text-Meaning Splitter)\n",
    "Uses embeddings or sentence similarity to decide where to break chunks.\n",
    "Prevents splitting mid-context if sentences are closely related.\n",
    "Example: Groups all related pain-assessment questions into one chunk."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###testingbelow\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !pip install langchain_experimental"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "\n",
    "# #might need to remove all together - don't think it's working\n",
    "# !pip install --upgrade langchain langchain-experimental\n",
    "# !pip install --upgrade langchain-community\n",
    "# !pip install langchain langchain-experimental langchain-community\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/tmp/ipykernel_76652/2495904805.py:7: LangChainDeprecationWarning: The class `HuggingFaceEmbeddings` was deprecated in LangChain 0.2.2 and will be removed in 1.0. An updated version of the class exists in the :class:`~langchain-huggingface package and should be used instead. To use it run `pip install -U :class:`~langchain-huggingface` and import as `from :class:`~langchain_huggingface import HuggingFaceEmbeddings``.\n",
      "  embedding_model = HuggingFaceEmbeddings(model_name=model_id)\n"
     ]
    }
   ],
   "source": [
    "\n",
    "\n",
    "from langchain_experimental.text_splitter import SemanticChunker\n",
    "\n",
    "from langchain.embeddings import HuggingFaceInferenceAPIEmbeddings\n",
    "\n",
    "from langchain.embeddings import HuggingFaceEmbeddings\n",
    "model_id = \"Snowflake/snowflake-arctic-embed-m\"\n",
    "embedding_model = HuggingFaceEmbeddings(model_name=model_id)\n",
    "\n",
    "semantic_splitter = SemanticChunker(embedding_model)\n",
    "\n",
    "all_documents = text_loader.load()\n",
    "documents_with_metadata = []\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain.schema import Document\n",
    "\n",
    "for doc in all_documents:\n",
    "    source_name = doc.metadata.get(\"source\", \"unknown\")  # Get document source\n",
    "\n",
    "    # Use SemanticChunker to intelligently split text\n",
    "    chunks = semantic_splitter.split_text(doc.page_content)\n",
    "\n",
    "    # Convert chunks into LangChain Document format with metadata\n",
    "    for chunk in chunks:\n",
    "        doc_chunk = Document(page_content=chunk, metadata={\"source\": source_name})\n",
    "        documents_with_metadata.append(doc_chunk)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##########################new testing below"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "#training_documents = text_loader.load()\n",
    "training_documents = documents_with_metadata"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "import uuid\n",
    "\n",
    "id_set = set()\n",
    "\n",
    "for document in training_documents:\n",
    "  id = str(uuid.uuid4())\n",
    "  while id in id_set:\n",
    "    id = uuid.uuid4()\n",
    "  id_set.add(id)\n",
    "  document.metadata[\"id\"] = id"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set: 4 docs\n",
      "Validation set: 1 docs\n",
      "Test set: 2 docs\n"
     ]
    }
   ],
   "source": [
    "# Define split percentages\n",
    "train_ratio = 0.7  # 70% training\n",
    "val_ratio = 0.2    # 20% validation\n",
    "test_ratio = 0.1   # 10% test\n",
    "\n",
    "# Calculate index breakpoints\n",
    "total_docs = len(training_documents)\n",
    "train_size = int(total_docs * train_ratio)\n",
    "val_size = int(total_docs * val_ratio)\n",
    "\n",
    "# Perform the splits\n",
    "training_split_documents = training_documents[:train_size]\n",
    "val_split_documents = training_documents[train_size:train_size + val_size]\n",
    "test_split_documents = training_documents[train_size + val_size:]\n",
    "\n",
    "# Print sizes to verify\n",
    "print(f\"Training set: {len(training_split_documents)} docs\")\n",
    "print(f\"Validation set: {len(val_split_documents)} docs\")\n",
    "print(f\"Test set: {len(test_split_documents)} docs\")\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain_openai import ChatOpenAI\n",
    "\n",
    "qa_chat_model = ChatOpenAI(\n",
    "    model=\"gpt-4o-mini\",\n",
    "    temperature=0\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain_core.prompts import ChatPromptTemplate\n",
    "\n",
    "qa_prompt = \"\"\"\\\n",
    "Given the following context, you must generate questions based on only the provided context.\n",
    "\n",
    "You are to generate {n_questions} questions which should be provided in the following format:\n",
    "\n",
    "1. QUESTION #1\n",
    "2. QUESTION #2\n",
    "...\n",
    "\n",
    "Context:\n",
    "{context}\n",
    "\"\"\"\n",
    "\n",
    "qa_prompt_template = ChatPromptTemplate.from_template(qa_prompt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "question_generation_chain = qa_prompt_template | qa_chat_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "import asyncio\n",
    "import uuid\n",
    "from tqdm import tqdm\n",
    "\n",
    "async def process_document(document, n_questions):\n",
    "    questions_generated = await question_generation_chain.ainvoke({\"context\": document.page_content, \"n_questions\": n_questions})\n",
    "\n",
    "    doc_questions = {}\n",
    "    doc_relevant_docs = {}\n",
    "\n",
    "    for question in questions_generated.content.split(\"\\n\"):\n",
    "        question_id = str(uuid.uuid4())\n",
    "        doc_questions[question_id] = \"\".join(question.split(\".\")[1:]).strip()\n",
    "        doc_relevant_docs[question_id] = [document.metadata[\"id\"]]\n",
    "\n",
    "    return doc_questions, doc_relevant_docs\n",
    "\n",
    "async def create_questions(documents, n_questions):\n",
    "    tasks = [process_document(doc, n_questions) for doc in documents]\n",
    "\n",
    "    questions = {}\n",
    "    relevant_docs = {}\n",
    "\n",
    "    for task in tqdm(asyncio.as_completed(tasks), total=len(documents), desc=\"Processing documents\"):\n",
    "        doc_questions, doc_relevant_docs = await task\n",
    "        questions.update(doc_questions)\n",
    "        relevant_docs.update(doc_relevant_docs)\n",
    "\n",
    "    return questions, relevant_docs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Processing documents: 100%|██████████| 4/4 [00:01<00:00,  3.75it/s]\n",
      "Processing documents: 100%|██████████| 1/1 [00:00<00:00,  1.21it/s]\n",
      "Processing documents: 100%|██████████| 2/2 [00:01<00:00,  1.98it/s]\n"
     ]
    }
   ],
   "source": [
    "training_questions, training_relevant_contexts = await create_questions(training_split_documents, 2)\n",
    "val_questions, val_relevant_contexts = await create_questions(val_split_documents, 2)\n",
    "test_questions, test_relevant_contexts = await create_questions(test_split_documents, 2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "training_corpus = {train_item.metadata[\"id\"] : train_item.page_content for train_item in training_split_documents}\n",
    "\n",
    "train_dataset = {\n",
    "    \"questions\" : training_questions,\n",
    "    \"relevant_contexts\" : training_relevant_contexts,\n",
    "    \"corpus\" : training_corpus\n",
    "}\n",
    "\n",
    "with open(\"training_dataset.jsonl\", \"w\") as f:\n",
    "  json.dump(train_dataset, f)\n",
    "\n",
    "\n",
    "val_corpus = {val_item.metadata[\"id\"] : val_item.page_content for val_item in val_split_documents}\n",
    "\n",
    "val_dataset = {\n",
    "    \"questions\" : val_questions,\n",
    "    \"relevant_contexts\" : val_relevant_contexts,\n",
    "    \"corpus\" : val_corpus\n",
    "}\n",
    "\n",
    "with open(\"val_dataset.jsonl\", \"w\") as f:\n",
    "  json.dump(val_dataset, f)\n",
    "\n",
    "\n",
    "train_corpus = {test_item.metadata[\"id\"] : test_item.page_content for test_item in test_split_documents}\n",
    "\n",
    "test_dataset = {\n",
    "    \"questions\" : test_questions,\n",
    "    \"relevant_contexts\" : test_relevant_contexts,\n",
    "    \"corpus\" : train_corpus\n",
    "}\n",
    "\n",
    "with open(\"test_dataset.jsonl\", \"w\") as f:\n",
    "  json.dump(test_dataset, f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !pip install -qU sentence_transformers datasets pyarrow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sentence_transformers import SentenceTransformer\n",
    "\n",
    "model_id = \"Snowflake/snowflake-arctic-embed-m\"\n",
    "model = SentenceTransformer(model_id)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.utils.data import DataLoader\n",
    "from torch.utils.data import Dataset\n",
    "from sentence_transformers import InputExample"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "BATCH_SIZE = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "corpus = train_dataset['corpus']\n",
    "queries = train_dataset['questions']\n",
    "relevant_docs = train_dataset['relevant_contexts']\n",
    "\n",
    "examples = []\n",
    "for query_id, query in queries.items():\n",
    "    doc_id = relevant_docs[query_id][0]\n",
    "    text = corpus[doc_id]\n",
    "    example = InputExample(texts=[query, text])\n",
    "    examples.append(example)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "loader = DataLoader(\n",
    "    examples, batch_size=BATCH_SIZE\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sentence_transformers.losses import MatryoshkaLoss, MultipleNegativesRankingLoss\n",
    "\n",
    "matryoshka_dimensions = [768, 512, 256, 128, 64]\n",
    "inner_train_loss = MultipleNegativesRankingLoss(model)\n",
    "train_loss = MatryoshkaLoss(\n",
    "    model, inner_train_loss, matryoshka_dims=matryoshka_dimensions\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sentence_transformers.evaluation import InformationRetrievalEvaluator\n",
    "\n",
    "corpus = val_dataset['corpus']\n",
    "queries = val_dataset['questions']\n",
    "relevant_docs = val_dataset['relevant_contexts']\n",
    "\n",
    "evaluator = InformationRetrievalEvaluator(queries, corpus, relevant_docs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "EPOCHS = 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<button onClick=\"this.nextSibling.style.display='block';this.style.display='none';\">Display W&B run</button><iframe src='https://wandb.ai/dummy/dummy/runs/bel6hiln?jupyter=true' style='border:none;width:100%;height:420px;display:none;'></iframe>"
      ],
      "text/plain": [
       "<wandb.sdk.wandb_run.Run at 0x72704850af90>"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#!pip install wandb\n",
    "\n",
    "import wandb\n",
    "wandb.init(mode=\"disabled\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# !pip install torch\n",
    "# !pip install accelerate>=0.26.0\n",
    "# !pip install transformers\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "#!pip install --upgrade --force-reinstall transformers accelerate torch\n",
    "#!which python\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "      \n",
       "      <progress value='5' max='5' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      [5/5 00:01, Epoch 5/5]\n",
       "    </div>\n",
       "    <table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       " <tr style=\"text-align: left;\">\n",
       "      <th>Step</th>\n",
       "      <th>Training Loss</th>\n",
       "      <th>Validation Loss</th>\n",
       "      <th>Cosine Accuracy@1</th>\n",
       "      <th>Cosine Accuracy@3</th>\n",
       "      <th>Cosine Accuracy@5</th>\n",
       "      <th>Cosine Accuracy@10</th>\n",
       "      <th>Cosine Precision@1</th>\n",
       "      <th>Cosine Precision@3</th>\n",
       "      <th>Cosine Precision@5</th>\n",
       "      <th>Cosine Precision@10</th>\n",
       "      <th>Cosine Recall@1</th>\n",
       "      <th>Cosine Recall@3</th>\n",
       "      <th>Cosine Recall@5</th>\n",
       "      <th>Cosine Recall@10</th>\n",
       "      <th>Cosine Ndcg@10</th>\n",
       "      <th>Cosine Mrr@10</th>\n",
       "      <th>Cosine Map@100</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>No log</td>\n",
       "      <td>No log</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.333333</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>No log</td>\n",
       "      <td>No log</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.333333</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>No log</td>\n",
       "      <td>No log</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.333333</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>No log</td>\n",
       "      <td>No log</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.333333</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>No log</td>\n",
       "      <td>No log</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.333333</td>\n",
       "      <td>0.200000</td>\n",
       "      <td>0.100000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table><p>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "warmup_steps = int(len(loader) * EPOCHS * 0.1)\n",
    "\n",
    "model.fit(\n",
    "    train_objectives=[(loader, train_loss)],\n",
    "    epochs=EPOCHS,\n",
    "    warmup_steps=warmup_steps,\n",
    "    output_path='models/midterm-compare-arctic-embed-m-ft',\n",
    "    show_progress_bar=True,\n",
    "    evaluator=evaluator,\n",
    "    evaluation_steps=50\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c3832f15349447c59ef0b7950d732a59",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "model.safetensors:   0%|          | 0.00/436M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "'https://huggingface.co/drewgenai/midterm-compare-arctic-embed-m-ft/commit/695a90e0d9d4a6ca560a5844c0e5a7cf4c4c74a9'"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.push_to_hub(f\"{hf_username}/midterm-compare-arctic-embed-m-ft\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5a84694a9cff451581d43a244cbd6ce5",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "modules.json:   0%|          | 0.00/349 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "d9635815ad784cc68833f2b4199c611b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "config_sentence_transformers.json:   0%|          | 0.00/281 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b425eef83f6c47cf90d9ad8df35bed07",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "README.md:   0%|          | 0.00/26.3k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "1c080b01bb4c43e3b0af3da190feff91",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "sentence_bert_config.json:   0%|          | 0.00/53.0 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "8ebbd4faaa99434fbd6413f24fadc8b1",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "config.json:   0%|          | 0.00/675 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "5ef43ded862f4e5685af4b66e51922af",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "model.safetensors:   0%|          | 0.00/436M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Some weights of BertModel were not initialized from the model checkpoint at drewgenai/midterm-compare-arctic-embed-m-ft and are newly initialized: ['pooler.dense.bias', 'pooler.dense.weight']\n",
      "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
     ]
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "f2704b3d8d214414acf54e23efb2de25",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "tokenizer_config.json:   0%|          | 0.00/1.41k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "70d0aca65df94b8c973d9e2aef700c6b",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "vocab.txt:   0%|          | 0.00/232k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "b8a288bc2740416d8be044c1534138a0",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "tokenizer.json:   0%|          | 0.00/712k [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "fd5494a1b2d2483884ccdfeaaf03e65c",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "special_tokens_map.json:   0%|          | 0.00/695 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e6259269b65b45358940c42ac8e9d127",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "1_Pooling%2Fconfig.json:   0%|          | 0.00/296 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "finetune_embeddings = HuggingFaceEmbeddings(model_name=f\"{hf_username}/midterm-compare-arctic-embed-m-ft\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###testingabove"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "#!pip install -qU huggingface_hub\n",
    "#!pip install -qU ipywidgets\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Some weights of BertModel were not initialized from the model checkpoint at drewgenai/demo-compare-arctic-embed-m-ft and are newly initialized: ['pooler.dense.bias', 'pooler.dense.weight']\n",
      "You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.\n"
     ]
    }
   ],
   "source": [
    "from sentence_transformers import SentenceTransformer\n",
    "from langchain.vectorstores import Qdrant\n",
    "from langchain.embeddings import HuggingFaceEmbeddings\n",
    "\n",
    "\n",
    "# Load the SentenceTransformer model\n",
    "#model_id = \"Snowflake/snowflake-arctic-embed-m\"\n",
    "model_id = f\"{hf_username}/demo-compare-arctic-embed-m-ft\"  \n",
    "embedding_model = HuggingFaceEmbeddings(model_name=model_id)\n",
    "# model_id = \"Snowflake/snowflake-arctic-embed-m\"\n",
    "# embedding_model = HuggingFaceEmbeddings(model_name=model_id)\n",
    "# model_id = \"Snowflake/snowflake-arctic-embed-m-v2.0\"\n",
    "# embedding_model = HuggingFaceEmbeddings(model_name=model_id, model_kwargs={\"trust_remote_code\": True})\n",
    "\n",
    "\n",
    "# Load documents into Qdrant\n",
    "qdrant_vectorstore = Qdrant.from_documents(\n",
    "    documents_with_metadata,\n",
    "    embedding_model,\n",
    "    location=\":memory:\",  # In-memory for testing\n",
    "    collection_name=\"document_comparison\",\n",
    ")\n",
    "\n",
    "# Create a retriever\n",
    "qdrant_retriever = qdrant_vectorstore.as_retriever()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "# from langchain_core.prompts import ChatPromptTemplate\n",
    "\n",
    "# RAG_PROMPT = \"\"\"\n",
    "# CONTEXT:\n",
    "# {context}\n",
    "\n",
    "# QUERY:\n",
    "# {question}\n",
    "\n",
    "# You are a helpful assistant. Use the available context to answer the question. If you can't answer the question, say you don't know.\n",
    "# \"\"\"\n",
    "\n",
    "# rag_prompt = ChatPromptTemplate.from_template(RAG_PROMPT)\n",
    "\n",
    "# from langchain_openai import ChatOpenAI\n",
    "\n",
    "# #openai_chat_model = ChatOpenAI(model=\"gpt-4o\")\n",
    "# openai_chat_model = ChatOpenAI(model=\"gpt-4o-mini\")\n",
    "\n",
    "# from operator import itemgetter\n",
    "# from langchain.schema.output_parser import StrOutputParser\n",
    "\n",
    "# rag_chain = (\n",
    "#     {\"context\": itemgetter(\"question\") | qdrant_retriever, \"question\": itemgetter(\"question\")}\n",
    "#     | rag_prompt | openai_chat_model | StrOutputParser()\n",
    "# )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain_core.prompts import ChatPromptTemplate\n",
    "RAG_PROMPT = \"\"\"\n",
    "CONTEXT:\n",
    "{context}\n",
    "\n",
    "QUERY:\n",
    "{question}\n",
    "\n",
    "You are a helpful assistant. Use the available context to answer the question.\n",
    "\n",
    "Return the response in **valid JSON format** with the following structure:\n",
    "\n",
    "[\n",
    "    {{\n",
    "        \"Derived Description\": \"A short name for the matched concept\",\n",
    "        \"Protocol_1_Name\": \"Protocol 1 - Matching Element\",\n",
    "        \"Protocol_2_Name\": \"Protocol 2 - Matching Element\"\n",
    "    }},\n",
    "    ...\n",
    "]\n",
    "\n",
    "### Rules:\n",
    "1. Only output **valid JSON** with no explanations, summaries, or markdown formatting.\n",
    "2. Ensure each entry in the JSON list represents a single matched data element from the two protocols.\n",
    "3. If no matching element is found in a protocol, leave it empty (\"\").\n",
    "4. **Do NOT include headers, explanations, or additional formatting**—only return the raw JSON list.\n",
    "5. It should include all the elements in the two protocols.\n",
    "6. If it cannot match the element, create the row and include the protocol it did find and put \"could not match\" in the other protocol column.\n",
    "\"\"\"\n",
    "\n",
    "rag_prompt = ChatPromptTemplate.from_template(RAG_PROMPT)\n",
    "\n",
    "from langchain_openai import ChatOpenAI\n",
    "\n",
    "#openai_chat_model = ChatOpenAI(model=\"gpt-4o\")\n",
    "openai_chat_model = ChatOpenAI(model=\"gpt-4o-mini\")\n",
    "\n",
    "from operator import itemgetter\n",
    "from langchain.schema.output_parser import StrOutputParser\n",
    "\n",
    "rag_chain = (\n",
    "    {\"context\": itemgetter(\"question\") | qdrant_retriever, \"question\": itemgetter(\"question\")}\n",
    "    | rag_prompt | openai_chat_model | StrOutputParser()\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "question_text = \"\"\"Between these two files containing protocols, can you find the data elements in each that most likely match the element in the other and output a CSV that lists three columns:\n",
    "\n",
    "The questions within elements will be similar between the two documents and can be used to match the elements.\n",
    "\n",
    "1. Derived description from the two documents describing the index/measure/scale.\n",
    "2. A column for each standard.\n",
    "3. In the column for each name/version, the data element used to capture that description.\n",
    "\n",
    "There should only be one row for each scale/index/etc.\n",
    "The description should not be one of the questions but a name that best describes the similar data elements.\"\"\"\n",
    "\n",
    "response_text = rag_chain.invoke({\"question\": question_text})\n",
    "# response = rag_chain.invoke({\"question\": question_text})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "✅ CSV file saved: matching_data_elements.csv\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import pandas as pd\n",
    "\n",
    "def parse_rag_output(response_text):\n",
    "    \"\"\"Extract structured JSON data from the RAG response.\"\"\"\n",
    "    try:\n",
    "        structured_data = json.loads(response_text)\n",
    "\n",
    "        # Ensure similarity score is always included\n",
    "        for item in structured_data:\n",
    "            item.setdefault(\"Similarity Score\", \"N/A\")  # Default if missing\n",
    "\n",
    "        return structured_data\n",
    "    except json.JSONDecodeError:\n",
    "        print(\"Error: Response is not valid JSON.\")\n",
    "        return None\n",
    "\n",
    "def save_to_csv(data, directory=\"./output\", filename=\"matching_data_elements.csv\"):\n",
    "    \"\"\"Save structured data to CSV.\"\"\"\n",
    "    if not data:\n",
    "        print(\"No data to save.\")\n",
    "        return\n",
    "\n",
    "    file_path = os.path.join(directory, filename)\n",
    "    df = pd.DataFrame(data, columns=[\"Derived Description\", \"Protocol_1_Name\", \"Protocol_2_Name\"])  # Ensure correct columns\n",
    "    df.to_csv(file_path, index=False)\n",
    "    print(f\"✅ CSV file saved: {filename}\")\n",
    "\n",
    "# Run the pipeline\n",
    "structured_output = parse_rag_output(response_text)\n",
    "save_to_csv(structured_output)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "# rag_chain.invoke({\"question\" : \"Based on the types of questions asked under each heading.  can you identify the headings in one document that most closely match the second document.  list them e.g   paincoping/doc1  painstrategy/doc2\"})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# rag_chain.invoke({\"question\" : \"Based on the types of questions asked under each heading.  can you identify the headings in one document that most closely match the second document.  list them e.g   paincoping/doc1  painstrategy/doc2. these are example headings not the ones in the actual documents.  just list the matches not the rational.  Can you list multiple matches?\"})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": ".venv",
   "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": "3.13.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}