jupyter/QA-PDF-Local.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "9f7b5878-d158-4423-97d3-9baf3b4cfe1a",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.11/site-packages/sentence_transformers/cross_encoder/CrossEncoder.py:13: TqdmExperimentalWarning: Using `tqdm.autonotebook.tqdm` in notebook mode. Use `tqdm.tqdm` instead to force console mode (e.g. in jupyter console)\n",
      "  from tqdm.autonotebook import tqdm, trange\n"
     ]
    }
   ],
   "source": [
    "import irisnative\n",
    "import os\n",
    "import sentence_transformers\n",
    "from langchain_text_splitters import RecursiveCharacterTextSplitter\n",
    "from langchain_community.document_loaders import PyPDFDirectoryLoader\n",
    "import numpy as np\n",
    "\n",
    "import torch\n",
    "from transformers import pipeline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "915cd2cb-059f-40b1-86b3-7ead09269b50",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Connected\n"
     ]
    }
   ],
   "source": [
    "connection_string = \"iris:1972/LLMRAG\"\n",
    "username = \"superuser\"\n",
    "password = \"SYS\"\n",
    "\n",
    "connectionIRIS = irisnative.createConnection(connection_string, username, password)\n",
    "cursorIRIS = connectionIRIS.cursor()\n",
    "print(\"Connected\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "d18d9dd6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "c9d42458c25b44cc88459c2cd7c4015f",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "config.json:   0%|          | 0.00/879 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e0407f7e9d0a451db9d897308682d974",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "model.safetensors:   0%|          | 0.00/1.11G [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "43bb2584acd14013acf181ee094a6267",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "tokenizer_config.json:   0%|          | 0.00/453 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "16bc2b1b6f6148a9b422ff2a9a683939",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "tokenizer.json:   0%|          | 0.00/16.3M [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "53eeb25199a54f8981e493267d92b9ff",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "added_tokens.json:   0%|          | 0.00/23.0 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "243abb14caf44a579c3343548de7e480",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "special_tokens_map.json:   0%|          | 0.00/173 [00:00<?, ?B/s]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "qa_model = pipeline(\"question-answering\", \"timpal0l/mdeberta-v3-base-squad2\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "3961787a-327b-4f8a-bd8e-a929ee686eee",
   "metadata": {},
   "outputs": [],
   "source": [
    "if not os.path.isdir('/app/data/model/'):\n",
    "    modelEmbedding = sentence_transformers.SentenceTransformer('sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2')            \n",
    "    modelEmbedding.save('/app/data/model/')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "944333c1",
   "metadata": {},
   "outputs": [],
   "source": [
    "text_splitter = RecursiveCharacterTextSplitter(\n",
    "    chunk_size = 700,\n",
    "    chunk_overlap  = 50,\n",
    ")\n",
    "path = \"/app/data\"\n",
    "loader = PyPDFDirectoryLoader(path)\n",
    "docs_before_split = loader.load()\n",
    "docs_after_split = text_splitter.split_documents(docs_before_split)\n",
    "modelEmbedding = sentence_transformers.SentenceTransformer(\"/app/data/model/\")\n",
    "for doc in docs_after_split:\n",
    "    embeddings = modelEmbedding.encode(doc.page_content, normalize_embeddings=True)\n",
    "    array = np.array(embeddings)\n",
    "    formatted_array = np.vectorize('{:.12f}'.format)(array)\n",
    "    parameters = []\n",
    "    parameters.append(doc.metadata['source'])\n",
    "    parameters.append(str(doc.page_content))\n",
    "    parameters.append(str(','.join(formatted_array)))\n",
    "    cursorIRIS.execute(\"INSERT INTO LLMRAG.DOCUMENTCHUNK (Document, Phrase, VectorizedPhrase) VALUES (?, ?, TO_VECTOR(?,DECIMAL))\", parameters)\n",
    "connectionIRIS.commit()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "0f4596f7-e4a6-464b-9086-91c5c16e6093",
   "metadata": {},
   "outputs": [],
   "source": [
    "literalQuestion = \"¿Qué medicamento puede tomar mi hijo de 2 años para bajar la fiebre?\"\n",
    "question = modelEmbedding.encode(literalQuestion, normalize_embeddings=True)\n",
    "array = np.array(question)\n",
    "formatted_array = np.vectorize('{:.12f}'.format)(array)\n",
    "parameterQuery = []\n",
    "parameterQuery.append(str(','.join(formatted_array)))\n",
    "cursorIRIS.execute(\"SELECT distinct(Document) FROM (SELECT VECTOR_DOT_PRODUCT(VectorizedPhrase, TO_VECTOR(?, DECIMAL)) AS similarity, Document FROM LLMRAG.DOCUMENTCHUNK) WHERE similarity > 0.6\", parameterQuery)\n",
    "similarityRows = cursorIRIS.fetchall()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "c96f0539-cfa1-4049-b498-3a592d1b1c6a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[Row(Document='/APP/DATA/PROSPECTO_69726.HTML.PDF')]"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "similarityRows"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "426e6299-8911-46a7-a83d-789b570434fe",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'score': 0.5216773748397827,\n",
       " 'start': 58,\n",
       " 'end': 101,\n",
       " 'answer': '\\nDalsy 40 mg/ml suspensión oral \\nibuprofeno'}"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "context = ''\n",
    "for similarityRow in similarityRows:\n",
    "    for doc in docs_before_split:\n",
    "        if similarityRow[0] == doc.metadata['source'].upper():\n",
    "            context = context +\"\".join(doc.page_content)\n",
    "\n",
    "qa_model(question = \"¿Cómo se llama el medicamento descrito en el prospecto?\", context = context)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "dae9ebab-79d7-41f4-af08-090f0ac22d61",
   "metadata": {},
   "outputs": [],
   "source": [
    "connectionIRIS.close()\n"
   ]
  }
 ],
 "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": "3.11.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}