notebook/pdf_loader2.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "673acfe6",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/manasvi/Github/RAG-Pipeline/.venv/lib/python3.12/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"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 2 PDF files to process\n",
      "\n",
      "Processing: NIPS-2017-attention-is-all-you-need-Paper.pdf\n",
      "  ✓ Loaded 11 pages\n",
      "\n",
      "Processing: 2310.05421v1.pdf\n",
      "  ✓ Loaded 4 pages\n",
      "\n",
      "Total documents loaded: 15\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "from langchain_community.document_loaders import PyPDFLoader, PyMuPDFLoader\n",
    "from langchain_text_splitters import RecursiveCharacterTextSplitter\n",
    "from pathlib import Path\n",
    "### Read all the pdf's inside the directory\n",
    "def process_all_pdfs(pdf_directory):\n",
    "    \"\"\"Process all PDF files in a directory\"\"\"\n",
    "    all_documents = []\n",
    "    pdf_dir = Path(pdf_directory)\n",
    "    \n",
    "    # Find all PDF files recursively\n",
    "    pdf_files = list(pdf_dir.glob(\"**/*.pdf\"))\n",
    "    \n",
    "    print(f\"Found {len(pdf_files)} PDF files to process\")\n",
    "    \n",
    "    for pdf_file in pdf_files:\n",
    "        print(f\"\\nProcessing: {pdf_file.name}\")\n",
    "        try:\n",
    "            loader = PyPDFLoader(str(pdf_file))\n",
    "            documents = loader.load()\n",
    "            \n",
    "            # Add source information to metadata\n",
    "            for doc in documents:\n",
    "                doc.metadata['source_file'] = pdf_file.name\n",
    "                doc.metadata['file_type'] = 'pdf'\n",
    "            \n",
    "            all_documents.extend(documents)\n",
    "            print(f\"  ✓ Loaded {len(documents)} pages\")\n",
    "            \n",
    "        except Exception as e:\n",
    "            print(f\"  ✗ Error: {e}\")\n",
    "    \n",
    "    print(f\"\\nTotal documents loaded: {len(all_documents)}\")\n",
    "    return all_documents\n",
    "\n",
    "# Process all PDFs in the data directory\n",
    "all_pdf_documents = process_all_pdfs(\"../data/pdf\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "3c788874",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Split 15 documents into 62 chunks\n",
      "\n",
      "Example chunk:\n",
      "Content: Attention Is All You Need\n",
      "Ashish Vaswani∗\n",
      "Google Brain\n",
      "avaswani@google.com\n",
      "Noam Shazeer∗\n",
      "Google Brain\n",
      "noam@google.com\n",
      "Niki Parmar∗\n",
      "Google Research\n",
      "nikip@google.com\n",
      "Jakob Uszkoreit∗\n",
      "Google Research\n",
      "usz...\n",
      "Metadata: {'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 0, 'page_label': '1', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 0, 'page_label': '1', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Attention Is All You Need\\nAshish Vaswani∗\\nGoogle Brain\\navaswani@google.com\\nNoam Shazeer∗\\nGoogle Brain\\nnoam@google.com\\nNiki Parmar∗\\nGoogle Research\\nnikip@google.com\\nJakob Uszkoreit∗\\nGoogle Research\\nusz@google.com\\nLlion Jones∗\\nGoogle Research\\nllion@google.com\\nAidan N. Gomez∗†\\nUniversity of Toronto\\naidan@cs.toronto.edu\\nŁukasz Kaiser ∗\\nGoogle Brain\\nlukaszkaiser@google.com\\nIllia Polosukhin∗‡\\nillia.polosukhin@gmail.com\\nAbstract\\nThe dominant sequence transduction models are based on complex recurrent or\\nconvolutional neural networks that include an encoder and a decoder. The best\\nperforming models also connect the encoder and decoder through an attention\\nmechanism. We propose a new simple network architecture, the Transformer,\\nbased solely on attention mechanisms, dispensing with recurrence and convolutions\\nentirely. Experiments on two machine translation tasks show these models to\\nbe superior in quality while being more parallelizable and requiring significantly'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 0, 'page_label': '1', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='entirely. Experiments on two machine translation tasks show these models to\\nbe superior in quality while being more parallelizable and requiring significantly\\nless time to train. Our model achieves 28.4 BLEU on the WMT 2014 English-\\nto-German translation task, improving over the existing best results, including\\nensembles, by over 2 BLEU. On the WMT 2014 English-to-French translation task,\\nour model establishes a new single-model state-of-the-art BLEU score of 41.0 after\\ntraining for 3.5 days on eight GPUs, a small fraction of the training costs of the\\nbest models from the literature.\\n1 Introduction\\nRecurrent neural networks, long short-term memory [12] and gated recurrent [7] neural networks\\nin particular, have been firmly established as state of the art approaches in sequence modeling and\\ntransduction problems such as language modeling and machine translation [ 29, 2, 5]. Numerous\\nefforts have since continued to push the boundaries of recurrent language models and encoder-decoder'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 0, 'page_label': '1', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='transduction problems such as language modeling and machine translation [ 29, 2, 5]. Numerous\\nefforts have since continued to push the boundaries of recurrent language models and encoder-decoder\\narchitectures [31, 21, 13].\\n∗Equal contribution. Listing order is random. Jakob proposed replacing RNNs with self-attention and started\\nthe effort to evaluate this idea. Ashish, with Illia, designed and implemented the first Transformer models and\\nhas been crucially involved in every aspect of this work. Noam proposed scaled dot-product attention, multi-head\\nattention and the parameter-free position representation and became the other person involved in nearly every\\ndetail. Niki designed, implemented, tuned and evaluated countless model variants in our original codebase and\\ntensor2tensor. Llion also experimented with novel model variants, was responsible for our initial codebase, and\\nefficient inference and visualizations. Lukasz and Aidan spent countless long days designing various parts of and'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 0, 'page_label': '1', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='efficient inference and visualizations. Lukasz and Aidan spent countless long days designing various parts of and\\nimplementing tensor2tensor, replacing our earlier codebase, greatly improving results and massively accelerating\\nour research.\\n†Work performed while at Google Brain.\\n‡Work performed while at Google Research.\\n31st Conference on Neural Information Processing Systems (NIPS 2017), Long Beach, CA, USA.'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 1, 'page_label': '2', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Recurrent models typically factor computation along the symbol positions of the input and output\\nsequences. Aligning the positions to steps in computation time, they generate a sequence of hidden\\nstates ht, as a function of the previous hidden state ht−1 and the input for position t. This inherently\\nsequential nature precludes parallelization within training examples, which becomes critical at longer\\nsequence lengths, as memory constraints limit batching across examples. Recent work has achieved\\nsignificant improvements in computational efficiency through factorization tricks [18] and conditional\\ncomputation [26], while also improving model performance in case of the latter. The fundamental\\nconstraint of sequential computation, however, remains.\\nAttention mechanisms have become an integral part of compelling sequence modeling and transduc-\\ntion models in various tasks, allowing modeling of dependencies without regard to their distance in'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 1, 'page_label': '2', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Attention mechanisms have become an integral part of compelling sequence modeling and transduc-\\ntion models in various tasks, allowing modeling of dependencies without regard to their distance in\\nthe input or output sequences [2, 16]. In all but a few cases [22], however, such attention mechanisms\\nare used in conjunction with a recurrent network.\\nIn this work we propose the Transformer, a model architecture eschewing recurrence and instead\\nrelying entirely on an attention mechanism to draw global dependencies between input and output.\\nThe Transformer allows for significantly more parallelization and can reach a new state of the art in\\ntranslation quality after being trained for as little as twelve hours on eight P100 GPUs.\\n2 Background\\nThe goal of reducing sequential computation also forms the foundation of the Extended Neural GPU\\n[20], ByteNet [15] and ConvS2S [8], all of which use convolutional neural networks as basic building'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 1, 'page_label': '2', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='The goal of reducing sequential computation also forms the foundation of the Extended Neural GPU\\n[20], ByteNet [15] and ConvS2S [8], all of which use convolutional neural networks as basic building\\nblock, computing hidden representations in parallel for all input and output positions. In these models,\\nthe number of operations required to relate signals from two arbitrary input or output positions grows\\nin the distance between positions, linearly for ConvS2S and logarithmically for ByteNet. This makes\\nit more difficult to learn dependencies between distant positions [ 11]. In the Transformer this is\\nreduced to a constant number of operations, albeit at the cost of reduced effective resolution due\\nto averaging attention-weighted positions, an effect we counteract with Multi-Head Attention as\\ndescribed in section 3.2.\\nSelf-attention, sometimes called intra-attention is an attention mechanism relating different positions'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 1, 'page_label': '2', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='described in section 3.2.\\nSelf-attention, sometimes called intra-attention is an attention mechanism relating different positions\\nof a single sequence in order to compute a representation of the sequence. Self-attention has been\\nused successfully in a variety of tasks including reading comprehension, abstractive summarization,\\ntextual entailment and learning task-independent sentence representations [4, 22, 23, 19].\\nEnd-to-end memory networks are based on a recurrent attention mechanism instead of sequence-\\naligned recurrence and have been shown to perform well on simple-language question answering and\\nlanguage modeling tasks [28].\\nTo the best of our knowledge, however, the Transformer is the first transduction model relying\\nentirely on self-attention to compute representations of its input and output without using sequence-\\naligned RNNs or convolution. In the following sections, we will describe the Transformer, motivate'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 1, 'page_label': '2', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='aligned RNNs or convolution. In the following sections, we will describe the Transformer, motivate\\nself-attention and discuss its advantages over models such as [14, 15] and [8].\\n3 Model Architecture\\nMost competitive neural sequence transduction models have an encoder-decoder structure [5, 2, 29].\\nHere, the encoder maps an input sequence of symbol representations (x1,...,x n) to a sequence\\nof continuous representations z = (z1,...,z n). Given z, the decoder then generates an output\\nsequence (y1,...,y m) of symbols one element at a time. At each step the model is auto-regressive\\n[9], consuming the previously generated symbols as additional input when generating the next.\\nThe Transformer follows this overall architecture using stacked self-attention and point-wise, fully\\nconnected layers for both the encoder and decoder, shown in the left and right halves of Figure 1,\\nrespectively.\\n3.1 Encoder and Decoder Stacks'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 1, 'page_label': '2', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='connected layers for both the encoder and decoder, shown in the left and right halves of Figure 1,\\nrespectively.\\n3.1 Encoder and Decoder Stacks\\nEncoder: The encoder is composed of a stack of N = 6 identical layers. Each layer has two\\nsub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position-\\n2'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 2, 'page_label': '3', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Figure 1: The Transformer - model architecture.\\nwise fully connected feed-forward network. We employ a residual connection [10] around each of\\nthe two sub-layers, followed by layer normalization [ 1]. That is, the output of each sub-layer is\\nLayerNorm(x+ Sublayer(x)), where Sublayer(x) is the function implemented by the sub-layer\\nitself. To facilitate these residual connections, all sub-layers in the model, as well as the embedding\\nlayers, produce outputs of dimension dmodel = 512.\\nDecoder: The decoder is also composed of a stack of N = 6identical layers. In addition to the two\\nsub-layers in each encoder layer, the decoder inserts a third sub-layer, which performs multi-head\\nattention over the output of the encoder stack. Similar to the encoder, we employ residual connections\\naround each of the sub-layers, followed by layer normalization. We also modify the self-attention\\nsub-layer in the decoder stack to prevent positions from attending to subsequent positions. This'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 2, 'page_label': '3', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='around each of the sub-layers, followed by layer normalization. We also modify the self-attention\\nsub-layer in the decoder stack to prevent positions from attending to subsequent positions. This\\nmasking, combined with fact that the output embeddings are offset by one position, ensures that the\\npredictions for position ican depend only on the known outputs at positions less than i.\\n3.2 Attention\\nAn attention function can be described as mapping a query and a set of key-value pairs to an output,\\nwhere the query, keys, values, and output are all vectors. The output is computed as a weighted sum\\nof the values, where the weight assigned to each value is computed by a compatibility function of the\\nquery with the corresponding key.\\n3.2.1 Scaled Dot-Product Attention\\nWe call our particular attention \"Scaled Dot-Product Attention\" (Figure 2). The input consists of\\nqueries and keys of dimension dk, and values of dimension dv. We compute the dot products of the\\n3'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 3, 'page_label': '4', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Scaled Dot-Product Attention\\n Multi-Head Attention\\nFigure 2: (left) Scaled Dot-Product Attention. (right) Multi-Head Attention consists of several\\nattention layers running in parallel.\\nquery with all keys, divide each by √dk, and apply a softmax function to obtain the weights on the\\nvalues.\\nIn practice, we compute the attention function on a set of queries simultaneously, packed together\\ninto a matrix Q. The keys and values are also packed together into matrices Kand V. We compute\\nthe matrix of outputs as:\\nAttention(Q,K,V ) = softmax(QKT\\n√dk\\n)V (1)\\nThe two most commonly used attention functions are additive attention [2], and dot-product (multi-\\nplicative) attention. Dot-product attention is identical to our algorithm, except for the scaling factor\\nof 1√dk\\n. Additive attention computes the compatibility function using a feed-forward network with\\na single hidden layer. While the two are similar in theoretical complexity, dot-product attention is'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 3, 'page_label': '4', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='of 1√dk\\n. Additive attention computes the compatibility function using a feed-forward network with\\na single hidden layer. While the two are similar in theoretical complexity, dot-product attention is\\nmuch faster and more space-efficient in practice, since it can be implemented using highly optimized\\nmatrix multiplication code.\\nWhile for small values of dk the two mechanisms perform similarly, additive attention outperforms\\ndot product attention without scaling for larger values of dk [3]. We suspect that for large values of\\ndk, the dot products grow large in magnitude, pushing the softmax function into regions where it has\\nextremely small gradients 4. To counteract this effect, we scale the dot products by 1√dk\\n.\\n3.2.2 Multi-Head Attention\\nInstead of performing a single attention function with dmodel-dimensional keys, values and queries,\\nwe found it beneficial to linearly project the queries, keys and values htimes with different, learned'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 3, 'page_label': '4', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='we found it beneficial to linearly project the queries, keys and values htimes with different, learned\\nlinear projections to dk, dk and dv dimensions, respectively. On each of these projected versions of\\nqueries, keys and values we then perform the attention function in parallel, yielding dv-dimensional\\noutput values. These are concatenated and once again projected, resulting in the final values, as\\ndepicted in Figure 2.\\nMulti-head attention allows the model to jointly attend to information from different representation\\nsubspaces at different positions. With a single attention head, averaging inhibits this.\\n4To illustrate why the dot products get large, assume that the components of q and k are independent random\\nvariables with mean 0 and variance 1. Then their dot product, q · k = ∑dk\\ni=1 qiki, has mean 0 and variance dk.\\n4'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 4, 'page_label': '5', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='MultiHead(Q,K,V ) = Concat(head1,..., headh)WO\\nwhere headi = Attention(QWQ\\ni ,KW K\\ni ,VW V\\ni )\\nWhere the projections are parameter matricesWQ\\ni ∈Rdmodel×dk , WK\\ni ∈Rdmodel×dk , WV\\ni ∈Rdmodel×dv\\nand WO ∈Rhdv×dmodel .\\nIn this work we employ h = 8 parallel attention layers, or heads. For each of these we use\\ndk = dv = dmodel/h= 64. Due to the reduced dimension of each head, the total computational cost\\nis similar to that of single-head attention with full dimensionality.\\n3.2.3 Applications of Attention in our Model\\nThe Transformer uses multi-head attention in three different ways:\\n• In \"encoder-decoder attention\" layers, the queries come from the previous decoder layer,\\nand the memory keys and values come from the output of the encoder. This allows every\\nposition in the decoder to attend over all positions in the input sequence. This mimics the\\ntypical encoder-decoder attention mechanisms in sequence-to-sequence models such as\\n[31, 2, 8].'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 4, 'page_label': '5', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='position in the decoder to attend over all positions in the input sequence. This mimics the\\ntypical encoder-decoder attention mechanisms in sequence-to-sequence models such as\\n[31, 2, 8].\\n• The encoder contains self-attention layers. In a self-attention layer all of the keys, values\\nand queries come from the same place, in this case, the output of the previous layer in the\\nencoder. Each position in the encoder can attend to all positions in the previous layer of the\\nencoder.\\n• Similarly, self-attention layers in the decoder allow each position in the decoder to attend to\\nall positions in the decoder up to and including that position. We need to prevent leftward\\ninformation flow in the decoder to preserve the auto-regressive property. We implement this\\ninside of scaled dot-product attention by masking out (setting to −∞) all values in the input\\nof the softmax which correspond to illegal connections. See Figure 2.\\n3.3 Position-wise Feed-Forward Networks'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 4, 'page_label': '5', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='of the softmax which correspond to illegal connections. See Figure 2.\\n3.3 Position-wise Feed-Forward Networks\\nIn addition to attention sub-layers, each of the layers in our encoder and decoder contains a fully\\nconnected feed-forward network, which is applied to each position separately and identically. This\\nconsists of two linear transformations with a ReLU activation in between.\\nFFN(x) = max(0,xW1 + b1)W2 + b2 (2)\\nWhile the linear transformations are the same across different positions, they use different parameters\\nfrom layer to layer. Another way of describing this is as two convolutions with kernel size 1.\\nThe dimensionality of input and output is dmodel = 512, and the inner-layer has dimensionality\\ndff = 2048.\\n3.4 Embeddings and Softmax\\nSimilarly to other sequence transduction models, we use learned embeddings to convert the input\\ntokens and output tokens to vectors of dimension dmodel. We also use the usual learned linear transfor-'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 4, 'page_label': '5', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Similarly to other sequence transduction models, we use learned embeddings to convert the input\\ntokens and output tokens to vectors of dimension dmodel. We also use the usual learned linear transfor-\\nmation and softmax function to convert the decoder output to predicted next-token probabilities. In\\nour model, we share the same weight matrix between the two embedding layers and the pre-softmax\\nlinear transformation, similar to [24]. In the embedding layers, we multiply those weights by √dmodel.\\n3.5 Positional Encoding\\nSince our model contains no recurrence and no convolution, in order for the model to make use of the\\norder of the sequence, we must inject some information about the relative or absolute position of the\\ntokens in the sequence. To this end, we add \"positional encodings\" to the input embeddings at the\\n5'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 5, 'page_label': '6', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Table 1: Maximum path lengths, per-layer complexity and minimum number of sequential operations\\nfor different layer types. nis the sequence length, dis the representation dimension, kis the kernel\\nsize of convolutions and rthe size of the neighborhood in restricted self-attention.\\nLayer Type Complexity per Layer Sequential Maximum Path Length\\nOperations\\nSelf-Attention O(n2 ·d) O(1) O(1)\\nRecurrent O(n·d2) O(n) O(n)\\nConvolutional O(k·n·d2) O(1) O(logk(n))\\nSelf-Attention (restricted) O(r·n·d) O(1) O(n/r)\\nbottoms of the encoder and decoder stacks. The positional encodings have the same dimension dmodel\\nas the embeddings, so that the two can be summed. There are many choices of positional encodings,\\nlearned and fixed [8].\\nIn this work, we use sine and cosine functions of different frequencies:\\nPE(pos,2i) = sin(pos/100002i/dmodel )\\nPE(pos,2i+1) = cos(pos/100002i/dmodel )\\nwhere posis the position and iis the dimension. That is, each dimension of the positional encoding'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 5, 'page_label': '6', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='PE(pos,2i) = sin(pos/100002i/dmodel )\\nPE(pos,2i+1) = cos(pos/100002i/dmodel )\\nwhere posis the position and iis the dimension. That is, each dimension of the positional encoding\\ncorresponds to a sinusoid. The wavelengths form a geometric progression from 2πto 10000 ·2π. We\\nchose this function because we hypothesized it would allow the model to easily learn to attend by\\nrelative positions, since for any fixed offset k, PEpos+k can be represented as a linear function of\\nPEpos.\\nWe also experimented with using learned positional embeddings [8] instead, and found that the two\\nversions produced nearly identical results (see Table 3 row (E)). We chose the sinusoidal version\\nbecause it may allow the model to extrapolate to sequence lengths longer than the ones encountered\\nduring training.\\n4 Why Self-Attention\\nIn this section we compare various aspects of self-attention layers to the recurrent and convolu-'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 5, 'page_label': '6', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='during training.\\n4 Why Self-Attention\\nIn this section we compare various aspects of self-attention layers to the recurrent and convolu-\\ntional layers commonly used for mapping one variable-length sequence of symbol representations\\n(x1,...,x n) to another sequence of equal length (z1,...,z n), with xi,zi ∈Rd, such as a hidden\\nlayer in a typical sequence transduction encoder or decoder. Motivating our use of self-attention we\\nconsider three desiderata.\\nOne is the total computational complexity per layer. Another is the amount of computation that can\\nbe parallelized, as measured by the minimum number of sequential operations required.\\nThe third is the path length between long-range dependencies in the network. Learning long-range\\ndependencies is a key challenge in many sequence transduction tasks. One key factor affecting the\\nability to learn such dependencies is the length of the paths forward and backward signals have to'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 5, 'page_label': '6', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='dependencies is a key challenge in many sequence transduction tasks. One key factor affecting the\\nability to learn such dependencies is the length of the paths forward and backward signals have to\\ntraverse in the network. The shorter these paths between any combination of positions in the input\\nand output sequences, the easier it is to learn long-range dependencies [11]. Hence we also compare\\nthe maximum path length between any two input and output positions in networks composed of the\\ndifferent layer types.\\nAs noted in Table 1, a self-attention layer connects all positions with a constant number of sequentially\\nexecuted operations, whereas a recurrent layer requires O(n) sequential operations. In terms of\\ncomputational complexity, self-attention layers are faster than recurrent layers when the sequence\\nlength n is smaller than the representation dimensionality d, which is most often the case with'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 5, 'page_label': '6', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='computational complexity, self-attention layers are faster than recurrent layers when the sequence\\nlength n is smaller than the representation dimensionality d, which is most often the case with\\nsentence representations used by state-of-the-art models in machine translations, such as word-piece\\n[31] and byte-pair [25] representations. To improve computational performance for tasks involving\\nvery long sequences, self-attention could be restricted to considering only a neighborhood of size rin\\n6'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 6, 'page_label': '7', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='the input sequence centered around the respective output position. This would increase the maximum\\npath length to O(n/r). We plan to investigate this approach further in future work.\\nA single convolutional layer with kernel width k<n does not connect all pairs of input and output\\npositions. Doing so requires a stack of O(n/k) convolutional layers in the case of contiguous kernels,\\nor O(logk(n)) in the case of dilated convolutions [ 15], increasing the length of the longest paths\\nbetween any two positions in the network. Convolutional layers are generally more expensive than\\nrecurrent layers, by a factor of k. Separable convolutions [ 6], however, decrease the complexity\\nconsiderably, to O(k·n·d+ n·d2). Even with k = n, however, the complexity of a separable\\nconvolution is equal to the combination of a self-attention layer and a point-wise feed-forward layer,\\nthe approach we take in our model.'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 6, 'page_label': '7', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='convolution is equal to the combination of a self-attention layer and a point-wise feed-forward layer,\\nthe approach we take in our model.\\nAs side benefit, self-attention could yield more interpretable models. We inspect attention distributions\\nfrom our models and present and discuss examples in the appendix. Not only do individual attention\\nheads clearly learn to perform different tasks, many appear to exhibit behavior related to the syntactic\\nand semantic structure of the sentences.\\n5 Training\\nThis section describes the training regime for our models.\\n5.1 Training Data and Batching\\nWe trained on the standard WMT 2014 English-German dataset consisting of about 4.5 million\\nsentence pairs. Sentences were encoded using byte-pair encoding [ 3], which has a shared source-\\ntarget vocabulary of about 37000 tokens. For English-French, we used the significantly larger WMT\\n2014 English-French dataset consisting of 36M sentences and split tokens into a 32000 word-piece'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 6, 'page_label': '7', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='target vocabulary of about 37000 tokens. For English-French, we used the significantly larger WMT\\n2014 English-French dataset consisting of 36M sentences and split tokens into a 32000 word-piece\\nvocabulary [31]. Sentence pairs were batched together by approximate sequence length. Each training\\nbatch contained a set of sentence pairs containing approximately 25000 source tokens and 25000\\ntarget tokens.\\n5.2 Hardware and Schedule\\nWe trained our models on one machine with 8 NVIDIA P100 GPUs. For our base models using\\nthe hyperparameters described throughout the paper, each training step took about 0.4 seconds. We\\ntrained the base models for a total of 100,000 steps or 12 hours. For our big models,(described on the\\nbottom line of table 3), step time was 1.0 seconds. The big models were trained for 300,000 steps\\n(3.5 days).\\n5.3 Optimizer\\nWe used the Adam optimizer [17] with β1 = 0.9, β2 = 0.98 and ϵ= 10−9. We varied the learning\\nrate over the course of training, according to the formula:'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 6, 'page_label': '7', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='(3.5 days).\\n5.3 Optimizer\\nWe used the Adam optimizer [17] with β1 = 0.9, β2 = 0.98 and ϵ= 10−9. We varied the learning\\nrate over the course of training, according to the formula:\\nlrate= d−0.5\\nmodel ·min(step_num−0.5,step_num·warmup_steps−1.5) (3)\\nThis corresponds to increasing the learning rate linearly for the first warmup_stepstraining steps,\\nand decreasing it thereafter proportionally to the inverse square root of the step number. We used\\nwarmup_steps= 4000.\\n5.4 Regularization\\nWe employ three types of regularization during training:\\nResidual Dropout We apply dropout [27] to the output of each sub-layer, before it is added to the\\nsub-layer input and normalized. In addition, we apply dropout to the sums of the embeddings and the\\npositional encodings in both the encoder and decoder stacks. For the base model, we use a rate of\\nPdrop = 0.1.\\n7'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 7, 'page_label': '8', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Table 2: The Transformer achieves better BLEU scores than previous state-of-the-art models on the\\nEnglish-to-German and English-to-French newstest2014 tests at a fraction of the training cost.\\nModel\\nBLEU Training Cost (FLOPs)\\nEN-DE EN-FR EN-DE EN-FR\\nByteNet [15] 23.75\\nDeep-Att + PosUnk [32] 39.2 1.0 ·1020\\nGNMT + RL [31] 24.6 39.92 2.3 ·1019 1.4 ·1020\\nConvS2S [8] 25.16 40.46 9.6 ·1018 1.5 ·1020\\nMoE [26] 26.03 40.56 2.0 ·1019 1.2 ·1020\\nDeep-Att + PosUnk Ensemble [32] 40.4 8.0 ·1020\\nGNMT + RL Ensemble [31] 26.30 41.16 1.8 ·1020 1.1 ·1021\\nConvS2S Ensemble [8] 26.36 41.29 7.7 ·1019 1.2 ·1021\\nTransformer (base model) 27.3 38.1 3.3 · 1018\\nTransformer (big) 28.4 41.0 2.3 ·1019\\nLabel Smoothing During training, we employed label smoothing of value ϵls = 0.1 [30]. This\\nhurts perplexity, as the model learns to be more unsure, but improves accuracy and BLEU score.\\n6 Results\\n6.1 Machine Translation\\nOn the WMT 2014 English-to-German translation task, the big transformer model (Transformer (big)'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 7, 'page_label': '8', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='6 Results\\n6.1 Machine Translation\\nOn the WMT 2014 English-to-German translation task, the big transformer model (Transformer (big)\\nin Table 2) outperforms the best previously reported models (including ensembles) by more than 2.0\\nBLEU, establishing a new state-of-the-art BLEU score of 28.4. The configuration of this model is\\nlisted in the bottom line of Table 3. Training took 3.5 days on 8 P100 GPUs. Even our base model\\nsurpasses all previously published models and ensembles, at a fraction of the training cost of any of\\nthe competitive models.\\nOn the WMT 2014 English-to-French translation task, our big model achieves a BLEU score of 41.0,\\noutperforming all of the previously published single models, at less than 1/4 the training cost of the\\nprevious state-of-the-art model. The Transformer (big) model trained for English-to-French used\\ndropout rate Pdrop = 0.1, instead of 0.3.\\nFor the base models, we used a single model obtained by averaging the last 5 checkpoints, which'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 7, 'page_label': '8', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='dropout rate Pdrop = 0.1, instead of 0.3.\\nFor the base models, we used a single model obtained by averaging the last 5 checkpoints, which\\nwere written at 10-minute intervals. For the big models, we averaged the last 20 checkpoints. We\\nused beam search with a beam size of 4 and length penalty α= 0.6 [31]. These hyperparameters\\nwere chosen after experimentation on the development set. We set the maximum output length during\\ninference to input length + 50, but terminate early when possible [31].\\nTable 2 summarizes our results and compares our translation quality and training costs to other model\\narchitectures from the literature. We estimate the number of floating point operations used to train a\\nmodel by multiplying the training time, the number of GPUs used, and an estimate of the sustained\\nsingle-precision floating-point capacity of each GPU 5.\\n6.2 Model Variations\\nTo evaluate the importance of different components of the Transformer, we varied our base model'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 7, 'page_label': '8', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='single-precision floating-point capacity of each GPU 5.\\n6.2 Model Variations\\nTo evaluate the importance of different components of the Transformer, we varied our base model\\nin different ways, measuring the change in performance on English-to-German translation on the\\ndevelopment set, newstest2013. We used beam search as described in the previous section, but no\\ncheckpoint averaging. We present these results in Table 3.\\nIn Table 3 rows (A), we vary the number of attention heads and the attention key and value dimensions,\\nkeeping the amount of computation constant, as described in Section 3.2.2. While single-head\\nattention is 0.9 BLEU worse than the best setting, quality also drops off with too many heads.\\n5We used values of 2.8, 3.7, 6.0 and 9.5 TFLOPS for K80, K40, M40 and P100, respectively.\\n8'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 8, 'page_label': '9', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Table 3: Variations on the Transformer architecture. Unlisted values are identical to those of the base\\nmodel. All metrics are on the English-to-German translation development set, newstest2013. Listed\\nperplexities are per-wordpiece, according to our byte-pair encoding, and should not be compared to\\nper-word perplexities.\\nN d model dff h d k dv Pdrop ϵls\\ntrain PPL BLEU params\\nsteps (dev) (dev) ×106\\nbase 6 512 2048 8 64 64 0.1 0.1 100K 4.92 25.8 65\\n(A)\\n1 512 512 5.29 24.9\\n4 128 128 5.00 25.5\\n16 32 32 4.91 25.8\\n32 16 16 5.01 25.4\\n(B) 16 5.16 25.1 58\\n32 5.01 25.4 60\\n(C)\\n2 6.11 23.7 36\\n4 5.19 25.3 50\\n8 4.88 25.5 80\\n256 32 32 5.75 24.5 28\\n1024 128 128 4.66 26.0 168\\n1024 5.12 25.4 53\\n4096 4.75 26.2 90\\n(D)\\n0.0 5.77 24.6\\n0.2 4.95 25.5\\n0.0 4.67 25.3\\n0.2 5.47 25.7\\n(E) positional embedding instead of sinusoids 4.92 25.7\\nbig 6 1024 4096 16 0.3 300K 4.33 26.4 213\\nIn Table 3 rows (B), we observe that reducing the attention key size dk hurts model quality. This'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 8, 'page_label': '9', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='(E) positional embedding instead of sinusoids 4.92 25.7\\nbig 6 1024 4096 16 0.3 300K 4.33 26.4 213\\nIn Table 3 rows (B), we observe that reducing the attention key size dk hurts model quality. This\\nsuggests that determining compatibility is not easy and that a more sophisticated compatibility\\nfunction than dot product may be beneficial. We further observe in rows (C) and (D) that, as expected,\\nbigger models are better, and dropout is very helpful in avoiding over-fitting. In row (E) we replace our\\nsinusoidal positional encoding with learned positional embeddings [8], and observe nearly identical\\nresults to the base model.\\n7 Conclusion\\nIn this work, we presented the Transformer, the first sequence transduction model based entirely on\\nattention, replacing the recurrent layers most commonly used in encoder-decoder architectures with\\nmulti-headed self-attention.\\nFor translation tasks, the Transformer can be trained significantly faster than architectures based'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 8, 'page_label': '9', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='multi-headed self-attention.\\nFor translation tasks, the Transformer can be trained significantly faster than architectures based\\non recurrent or convolutional layers. On both WMT 2014 English-to-German and WMT 2014\\nEnglish-to-French translation tasks, we achieve a new state of the art. In the former task our best\\nmodel outperforms even all previously reported ensembles.\\nWe are excited about the future of attention-based models and plan to apply them to other tasks. We\\nplan to extend the Transformer to problems involving input and output modalities other than text and\\nto investigate local, restricted attention mechanisms to efficiently handle large inputs and outputs\\nsuch as images, audio and video. Making generation less sequential is another research goals of ours.\\nThe code we used to train and evaluate our models is available at https://github.com/\\ntensorflow/tensor2tensor.\\nAcknowledgements We are grateful to Nal Kalchbrenner and Stephan Gouws for their fruitful'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 8, 'page_label': '9', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='tensorflow/tensor2tensor.\\nAcknowledgements We are grateful to Nal Kalchbrenner and Stephan Gouws for their fruitful\\ncomments, corrections and inspiration.\\n9'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 9, 'page_label': '10', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='References\\n[1] Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization. arXiv preprint\\narXiv:1607.06450, 2016.\\n[2] Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly\\nlearning to align and translate. CoRR, abs/1409.0473, 2014.\\n[3] Denny Britz, Anna Goldie, Minh-Thang Luong, and Quoc V . Le. Massive exploration of neural\\nmachine translation architectures. CoRR, abs/1703.03906, 2017.\\n[4] Jianpeng Cheng, Li Dong, and Mirella Lapata. Long short-term memory-networks for machine\\nreading. arXiv preprint arXiv:1601.06733, 2016.\\n[5] Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Fethi Bougares, Holger Schwenk,\\nand Yoshua Bengio. Learning phrase representations using rnn encoder-decoder for statistical\\nmachine translation. CoRR, abs/1406.1078, 2014.\\n[6] Francois Chollet. Xception: Deep learning with depthwise separable convolutions. arXiv\\npreprint arXiv:1610.02357, 2016.'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 9, 'page_label': '10', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='machine translation. CoRR, abs/1406.1078, 2014.\\n[6] Francois Chollet. Xception: Deep learning with depthwise separable convolutions. arXiv\\npreprint arXiv:1610.02357, 2016.\\n[7] Junyoung Chung, Çaglar Gülçehre, Kyunghyun Cho, and Yoshua Bengio. Empirical evaluation\\nof gated recurrent neural networks on sequence modeling. CoRR, abs/1412.3555, 2014.\\n[8] Jonas Gehring, Michael Auli, David Grangier, Denis Yarats, and Yann N. Dauphin. Convolu-\\ntional sequence to sequence learning. arXiv preprint arXiv:1705.03122v2, 2017.\\n[9] Alex Graves. Generating sequences with recurrent neural networks. arXiv preprint\\narXiv:1308.0850, 2013.\\n[10] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for im-\\nage recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern\\nRecognition, pages 770–778, 2016.\\n[11] Sepp Hochreiter, Yoshua Bengio, Paolo Frasconi, and Jürgen Schmidhuber. Gradient flow in'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 9, 'page_label': '10', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='Recognition, pages 770–778, 2016.\\n[11] Sepp Hochreiter, Yoshua Bengio, Paolo Frasconi, and Jürgen Schmidhuber. Gradient flow in\\nrecurrent nets: the difficulty of learning long-term dependencies, 2001.\\n[12] Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural computation,\\n9(8):1735–1780, 1997.\\n[13] Rafal Jozefowicz, Oriol Vinyals, Mike Schuster, Noam Shazeer, and Yonghui Wu. Exploring\\nthe limits of language modeling. arXiv preprint arXiv:1602.02410, 2016.\\n[14] Łukasz Kaiser and Ilya Sutskever. Neural GPUs learn algorithms. In International Conference\\non Learning Representations (ICLR), 2016.\\n[15] Nal Kalchbrenner, Lasse Espeholt, Karen Simonyan, Aaron van den Oord, Alex Graves, and Ko-\\nray Kavukcuoglu. Neural machine translation in linear time.arXiv preprint arXiv:1610.10099v2,\\n2017.\\n[16] Yoon Kim, Carl Denton, Luong Hoang, and Alexander M. Rush. Structured attention networks.\\nIn International Conference on Learning Representations, 2017.'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 9, 'page_label': '10', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='2017.\\n[16] Yoon Kim, Carl Denton, Luong Hoang, and Alexander M. Rush. Structured attention networks.\\nIn International Conference on Learning Representations, 2017.\\n[17] Diederik Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In ICLR, 2015.\\n[18] Oleksii Kuchaiev and Boris Ginsburg. Factorization tricks for LSTM networks. arXiv preprint\\narXiv:1703.10722, 2017.\\n[19] Zhouhan Lin, Minwei Feng, Cicero Nogueira dos Santos, Mo Yu, Bing Xiang, Bowen\\nZhou, and Yoshua Bengio. A structured self-attentive sentence embedding. arXiv preprint\\narXiv:1703.03130, 2017.\\n[20] Samy Bengio Łukasz Kaiser. Can active memory replace attention? In Advances in Neural\\nInformation Processing Systems, (NIPS), 2016.\\n10'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 10, 'page_label': '11', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='[21] Minh-Thang Luong, Hieu Pham, and Christopher D Manning. Effective approaches to attention-\\nbased neural machine translation. arXiv preprint arXiv:1508.04025, 2015.\\n[22] Ankur Parikh, Oscar Täckström, Dipanjan Das, and Jakob Uszkoreit. A decomposable attention\\nmodel. In Empirical Methods in Natural Language Processing, 2016.\\n[23] Romain Paulus, Caiming Xiong, and Richard Socher. A deep reinforced model for abstractive\\nsummarization. arXiv preprint arXiv:1705.04304, 2017.\\n[24] Ofir Press and Lior Wolf. Using the output embedding to improve language models. arXiv\\npreprint arXiv:1608.05859, 2016.\\n[25] Rico Sennrich, Barry Haddow, and Alexandra Birch. Neural machine translation of rare words\\nwith subword units. arXiv preprint arXiv:1508.07909, 2015.\\n[26] Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton,\\nand Jeff Dean. Outrageously large neural networks: The sparsely-gated mixture-of-experts\\nlayer. arXiv preprint arXiv:1701.06538, 2017.'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 10, 'page_label': '11', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='and Jeff Dean. Outrageously large neural networks: The sparsely-gated mixture-of-experts\\nlayer. arXiv preprint arXiv:1701.06538, 2017.\\n[27] Nitish Srivastava, Geoffrey E Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdi-\\nnov. Dropout: a simple way to prevent neural networks from overfitting. Journal of Machine\\nLearning Research, 15(1):1929–1958, 2014.\\n[28] Sainbayar Sukhbaatar, arthur szlam, Jason Weston, and Rob Fergus. End-to-end memory\\nnetworks. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett, editors,\\nAdvances in Neural Information Processing Systems 28, pages 2440–2448. Curran Associates,\\nInc., 2015.\\n[29] Ilya Sutskever, Oriol Vinyals, and Quoc VV Le. Sequence to sequence learning with neural\\nnetworks. In Advances in Neural Information Processing Systems, pages 3104–3112, 2014.\\n[30] Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna.'),\n",
       " Document(metadata={'producer': 'PyPDF2', 'creator': 'PyPDF', 'creationdate': '', 'subject': 'Neural Information Processing Systems http://nips.cc/', 'publisher': 'Curran Associates, Inc.', 'language': 'en-US', 'created': '2017', 'eventtype': 'Poster', 'description-abstract': 'The dominant sequence transduction models are based on complex recurrent orconvolutional neural networks in an encoder and decoder configuration. The best performing such models also connect the encoder and decoder through an attentionm echanisms.  We propose a novel, simple network architecture based solely onan attention mechanism, dispensing with recurrence and convolutions entirely.Experiments on two machine translation tasks show these models to be superiorin quality while being more parallelizable and requiring significantly less timeto train. Our single model with 165 million parameters, achieves 27.5 BLEU onEnglish-to-German translation, improving over the existing best ensemble result by over 1 BLEU. On English-to-French translation, we outperform the previoussingle state-of-the-art with model by 0.7 BLEU, achieving a BLEU score of 41.1.', 'title': 'Attention is All you Need', 'date': '2017', 'moddate': '2018-02-12T21:22:10-08:00', 'published': '2017', 'type': 'Conference Proceedings', 'firstpage': '5998', 'book': 'Advances in Neural Information Processing Systems 30', 'description': 'Paper accepted and presented at the Neural Information Processing Systems Conference (http://nips.cc/)', 'editors': 'I. Guyon and U.V. Luxburg and S. Bengio and H. Wallach and R. Fergus and S. Vishwanathan and R. Garnett', 'author': 'Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, Illia Polosukhin', 'lastpage': '6008', 'source': '../data/pdf/NIPS-2017-attention-is-all-you-need-Paper.pdf', 'total_pages': 11, 'page': 10, 'page_label': '11', 'source_file': 'NIPS-2017-attention-is-all-you-need-Paper.pdf', 'file_type': 'pdf'}, page_content='networks. In Advances in Neural Information Processing Systems, pages 3104–3112, 2014.\\n[30] Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jonathon Shlens, and Zbigniew Wojna.\\nRethinking the inception architecture for computer vision. CoRR, abs/1512.00567, 2015.\\n[31] Yonghui Wu, Mike Schuster, Zhifeng Chen, Quoc V Le, Mohammad Norouzi, Wolfgang\\nMacherey, Maxim Krikun, Yuan Cao, Qin Gao, Klaus Macherey, et al. Google’s neural machine\\ntranslation system: Bridging the gap between human and machine translation. arXiv preprint\\narXiv:1609.08144, 2016.\\n[32] Jie Zhou, Ying Cao, Xuguang Wang, Peng Li, and Wei Xu. Deep recurrent models with\\nfast-forward connections for neural machine translation. CoRR, abs/1606.04199, 2016.\\n11'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 0, 'page_label': '1', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='Submitted to the 3rd International Conference on “Women in Science & Technology: Creating Sustainable Career”  \\n28 -30 December, 2023 \\nAutomating Customer Service using LangChain \\nBuilding custom open-source GPT Chatbot for organizations \\nKeivalya Pandya \\n19me439@bvmengineering.ac.in \\nBirla Vishvakarma Mahavidyalaya, Gujarat, India \\nProf. Dr. Mehfuza Holia \\nmsholia@bvmengineering.ac.in \\nBirla Vishvakarma Mahavidyalaya, Gujarat, India \\n \\nAbstract— In the digital age, the dynamics of customer \\nservice are evolving, driven by technological advancements and \\nthe integration of Large Language Models (LLMs). This research \\npaper introduces a groundbreaking approach to automating \\ncustomer service using LangChain, a custom LLM tailored for \\norganizations. The paper explores the obsolescence of traditional \\ncustomer support techniques, particul arly Frequently Asked \\nQuestions (FAQs), and proposes a paradigm shift towards \\nresponsive, context -aware, and personalized customer'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 0, 'page_label': '1', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='customer support techniques, particul arly Frequently Asked \\nQuestions (FAQs), and proposes a paradigm shift towards \\nresponsive, context -aware, and personalized customer \\ninteractions. The heart of this innovation lies in the fusion of \\nopen-source methodologies, web scraping, fine -tuning, and th e \\nseamless integration of LangCh ain into customer service \\nplatforms. This open -source state-of-the-art framework, \\npresented as \"Sahaay,\" demonstrates the ability to scale across \\nindustries and organizations, offering real -time support and \\nquery resolution. Key  elements of this research encompass data \\ncollection via web scraping, the role of embeddings, the \\nutilization of Google\\'s Flan T5 XXL , Base and S mall language \\nmodels for knowledge retrieval, and the integration of the \\nchatbot into customer service platfor ms. The results s ection \\nprovides i nsights into the ir performance and use cases, here \\nparticularly within an educational institution.  This research'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 0, 'page_label': '1', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='chatbot into customer service platfor ms. The results s ection \\nprovides i nsights into the ir performance and use cases, here \\nparticularly within an educational institution.  This research \\nheralds a new era in customer service, where technology is \\nharnessed to create effici ent, personalized, and r esponsive \\ninteractions. Sahaay, power ed b y LangChain, redefines the \\ncustomer-company relationship, elevating customer retention, \\nvalue extraction, and brand image. As organizations embrace \\nLLMs, customer service bec omes a dynamic and customer -\\ncentric ecosystem. \\nKeywords— Customer Service Automation, Large L anguage \\nModels, LangChain, Web Scraping, Context-Aware Interactions \\nI.  INTRODUCTION \\n“Customer is king” is the ancient  mantra reflecting the \\nsignificance of customers in every business. In the digital age, \\nwhere the rhythms of mode rn life are guided by the pu lse of \\ntechnology, the realm of customer service stands as the'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 0, 'page_label': '1', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='significance of customers in every business. In the digital age, \\nwhere the rhythms of mode rn life are guided by the pu lse of \\ntechnology, the realm of customer service stands as the \\nfrontline of engagement between businesses and their clientele. \\nIt is the place where queries are an swered, problems are \\nresolved, and trust is forged. \\nThis research paper brings the future of customer service, \\nwhere automation, personalization, and responsiveness \\nconverge to redefine the customer -company r elationship. At \\nthe heart of this transformation  lies the integration of LLMs, \\nexemplified by LangChain [1]. \\nIn t he annal s of customer service history, FAQs and \\ntraditional support mechanisms have long held sway. These \\nvenerable tools have dutifully served as repositories of \\ninformation, attempting to address the queries and concerns of \\ncustomers. However, as we stan d at the cusp of a new era in \\ncustomer service automation, it becomes abundantly clear that'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 0, 'page_label': '1', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='information, attempting to address the queries and concerns of \\ncustomers. However, as we stan d at the cusp of a new era in \\ncustomer service automation, it becomes abundantly clear that \\nthe traditional methods once hailed as revolutionary, are \\ngradually becoming obsolete. \\nThis paper i s an invitation to envision a future wher e \\ncustomer service is no t a cost center but a wellspring of \\ncustomer satisfaction and loyalty. We propose an open -source \\nframework that can be scaled to any industry or organization to \\nfulfill the consumer needs for  support and query resolution \\nwithin seconds. \\nFor demonstration p urposes, we use the information \\npresented by Birla Vishvakarma Mahavidyalaya (BVM) \\nEngineering College on their website \\nhttps://bvmengineering.ac.in/ as the context for our chatbot, \\nfrom where it can retrieve al l the information in real -time and \\nanswer to any queries that are raised by the users. Here, users \\ncan be anyone ranging from prospective  students, current'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 0, 'page_label': '1', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content=\"from where it can retrieve al l the information in real -time and \\nanswer to any queries that are raised by the users. Here, users \\ncan be anyone ranging from prospective  students, current \\nstudents who intend to get informat ion from the Notice Board, \\nresearchers wh o w ish to search for the ir potential research \\nguide, and so on. The applications are endless. \\nII. LITERATURE SURVEY \\nS. Kim (2023) et al addresses the challenge of deploying \\nresource-intensive large neural models, such a s Transformers, \\nfor information retrieval  (IR) while maintaining efficiency. \\nExperimental results on MSMARCO benchmarks demonstrate \\nthe effectiveness o f this approach, achieving successful \\ndistillation of both dual -encoder and cross -encoder teacher \\nmodels into smaller, 1/10th size asymmetric stud ents while \\nretaining 95-97% of the teacher's performance [2]. L. Bonifacio \\net al (2022) highlights the recent transformation in the\"),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 0, 'page_label': '1', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content=\"models into smaller, 1/10th size asymmetric stud ents while \\nretaining 95-97% of the teacher's performance [2]. L. Bonifacio \\net al (2022) highlights the recent transformation in the \\nInformation Retrieval (IR) field, propelled by the emergence of \\nlarge pretrained transformer models. The MS MARCO dataset \\nplayed a pivotal role in t his revolution, enabling zero -shot \\ntransfer learning across various tasks [3]. \\nThis paper proposed a novel open-source ap proach to \\nbuilding LLM Chatbots using custom knowledge from the \\ncontent in the website. It is unique in several ways: \\n1. We propose an open-source framework which is robust \\nwith the type of dataset available on the webpage or \\nthe web of links. \\n2. This implementation aims to compliment the use of \\nFAQs with a more interactive and us er-friendly \\ninterface. \\n3. We then do a c omparative study of various models, \\ntheir performance on  the provided data relative to the \\nexpected response from the LLM.\"),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 1, 'page_label': '2', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='Submitted to the 3rd International Conference on “Women in Science & Technology: Creating Sustainable Career”  \\n28 -30 December, 2023 \\nIII. METHODOLOGY \\nThis sect ion covers the data c ollection, details about the \\nselected model, fine -tuning, and integration with the Gradio \\nAPIs for web deployment. \\nA. Data Collection \\nTo gather the necessary data for our project, we employed \\nBeautifulSoup web scraping techniques to retri eve publicly \\naccessible information from an organization’s homepage. We \\nobserved this page is often linke d with all  the relevant \\ninformation required for the user/visitor. This approach \\nallowed us to collect a wide array of data, including customer \\nservice FAQs, product manual s, support forums, chat logs, \\nassociated institutions, and so on. This data further serves as \\nthe context for our LLM. \\nB. Embeddings \\nEmbeddings play a pivotal role in the development of any \\nLLM powered. They are vector representation s of words or'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 1, 'page_label': '2', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='the context for our LLM. \\nB. Embeddings \\nEmbeddings play a pivotal role in the development of any \\nLLM powered. They are vector representation s of words or \\nphrases in a continuous mathematical space that capture \\nsemantic a nd contextual information, allowing the model to \\nunderstand the meaning and re lationships between words, \\nwhich is essential for pro viding meaningful responses to user \\nqueries. \\nWe have used HuggingFace Instuct Embeddings – \\n“hkunlp/instructor-large” a text e mbedding model fine -tuned \\nfor specific tasks and domains, such as classification, retrieval, \\nclustering, and text evaluation [4]. What sets Instructor apart is \\nits ability to gener ate tailored text embeddings without \\nrequiring additional fine -tuning. These embeddings are then \\nstored using FAISS  (Facebook AI  Similarity Search)  library \\nthat allows developers to quickly search for embedd ings of \\nmultimedia documents that are similar to each other [5]. \\nC. Language Model'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 1, 'page_label': '2', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='that allows developers to quickly search for embedd ings of \\nmultimedia documents that are similar to each other [5]. \\nC. Language Model \\nWe have chosen Google’s Flan T5 XXL as the mo st \\nappropriate language model after comparing with other Flan T5 \\ndistributions to retrieve knowledge from the vectorspace and \\nchat_history (or memory) [6]. The model retains the context of \\nprevious messages  and uses that as a reference to predict \\nanswers for the upcoming questions. This helps us ers to have \\nan interactive conversation with the chatbot, instead of a \\nmonotonous and robotic one. \\nD. Integration with Customer Service Platforms \\nA simple chat window can be activated at the corner of any \\nwebsite which would enable users to interact with the chatbot \\nand ask any relevant questions or doubts regarding the \\norganization. However, for the d emonstration purpose of this \\npaper, we are using Gradio API framework [7]. \\nIV. RESULTS \\nIn this section, we mention the metric s of comparison,'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 1, 'page_label': '2', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='organization. However, for the d emonstration purpose of this \\npaper, we are using Gradio API framework [7]. \\nIV. RESULTS \\nIn this section, we mention the metric s of comparison, \\nprovide comparative analysis, and use cases in association with \\nan educational institution. \\nA. Evaluating the Performance of LLMs \\nIt is relatively difficul t to evaluate LangChain agents, \\nespecially when trained on large chunks of context data sets for \\ninformation retrieval. Hence, the current solution for the lack of \\nmetrics is to rely on human knowledge to get a sense of how \\nthe chain/agent is performing. \\nIt is evident from TABLE-I, II and III that the XXL model \\noutperforms other competitive LLMs such as BASE and \\nSMALL. \\n \\nFig. 1. Model Architecture – Sahaay'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 2, 'page_label': '3', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='Submitted to the 3rd International Conference on “Women in Science & Technology: Creating Sustainable Career”  \\n28 -30 December, 2023 \\nTABLE I.  Google’s Flan-T5-XXL Performance \\nMetrics \\nPerformance Sr. \\nNo. Query/Prompt Answer \\n1. What is BVM? Birla Vishvakarma \\nMahavidyalaya \\n✯✯✯✯ \\n2. Where is it?a Vallabh Vidyanagar, Gujarat \\nIndia \\n✯✯✯ \\n3. What is IEEE \\nBVM? \\nInstitute of Electrical and \\nElectronics Engineers Stu dent \\nBranch of BVM \\n✯✯✯✯ \\n4. What is TRS \\nBVM? \\nBVM Student Chapter \\nassociated with Rob otics \\nSociety India \\n✯✯✯✯ \\n5. Name the annual \\nnewsletter? \\nVishvakarma Magazine and \\nNewsletter \\n✯✯✯✯ \\n6. \\nTell me \\nsomething about \\nICWSTCSC \\nThe International Conference \\nfor Wome n in Science and \\nTechnology Creating  \\nSustainable Car eer is the 3rd \\nInternational Con ference \\nhappening in hybrid mode. \\n✯✯✯✯✯ \\na. Follow-up question \\nTABLE II.  GOOGLE’S FLAN-T5-BASE PERFORMANCE \\nMetrics \\nPerformance Sr. \\nNo. Query/Prompt Answer \\n1. What is BVM? \\nBVM is the first engineering'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 2, 'page_label': '3', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='✯✯✯✯✯ \\na. Follow-up question \\nTABLE II.  GOOGLE’S FLAN-T5-BASE PERFORMANCE \\nMetrics \\nPerformance Sr. \\nNo. Query/Prompt Answer \\n1. What is BVM? \\nBVM is the first engineering \\ncollege of Gujarat State \\nestablished way back in 1948. \\n✯✯✯✯ \\n2. Where is it?b VV Nagar, Gujarat, India ✯✯✯ \\n3. What is IEEE \\nBVM? \\nfirst engineering college of \\nGujarat State established way \\nback in 1948 \\n✯ \\n4. What is TRS \\nBVM? \\nCVM BVM Roboti cs Society \\nOf India \\n✯✯ \\n5. Name the annual \\nnewsletter? \\nVishvakarma Magazine ✯✯✯✯ \\n6. \\nTell me \\nsomething about \\nICWSTCSC \\nBVM 3rd Internartional \\nConference \\n✯✯✯✯ \\nb. Follow-up question \\nTABLE III.  GOOGLE’S FLAN-T5-SMALL PERFORMANCE \\nMetrics \\nPerformance Sr. \\nNo. Query/Prompt Answer \\n1. What is BVM? \\nBVM is the first Autonomous \\nEngineering institute of Gujarat \\nto obtain academic autonomy \\nfor all its UG & PG programs, \\nfrom University Grant \\nCommission (UGC). \\n✯✯✯ \\n2. Where is it?c BVM Engineering College ✯ \\n3. What is IEEE \\nBVM? \\nengineering college of Gujarat'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 2, 'page_label': '3', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content=\"for all its UG & PG programs, \\nfrom University Grant \\nCommission (UGC). \\n✯✯✯ \\n2. Where is it?c BVM Engineering College ✯ \\n3. What is IEEE \\nBVM? \\nengineering college of Gujarat \\nState \\n✯ \\n4. What is TRS \\nBVM? \\nBVM ✯ \\n5. Name the annual \\nnewsletter? \\nCampus Publications ✯ \\n6. \\nTell me \\nsomething about \\nICWSTCSC \\nICWSTCSC 2023 PMSSS \\nstudents' reporting after \\nadmission (AY: 2023 -24) at \\nBVM 3rd Internartional \\nConference. \\n✯✯✯ \\nc. Follow-up question \\nB. Applications \\nCustomer service establishes a direct connection between \\nthe customer and t he company. It  retains customers and \\nextracts higher value from t hem. By harnessing the power of \\nLarge Language Models  as shown in  Fig. 2, customer service \\ncan be elevate d to new heights, facilitating efficient, \\npersonalized, and responsive interactions. The LangChain fine-\\ntuned over custom knowledge of the product, service, or \\norganization can effectively address a wide array of customer\"),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 2, 'page_label': '3', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='personalized, and responsive interactions. The LangChain fine-\\ntuned over custom knowledge of the product, service, or \\norganization can effectively address a wide array of customer \\ninquiries and issues. Its ability to understand context and \\nhistory empowers it to provide personalized support to \\ncustomers. Automated customer service powered by LLMs is \\navailable around the clock and i s also proficient in multiple \\nlanguages. \\nV. CONCLUSION \\nIn the ever -evolving landscape of  customer service, the \\nintroduction of Sahaay’s innovative approach presented in this \\npaper, using LangChain as a prime example, ushered in a new \\nera of automation.  Automating customer se rvice using \\nSahaay’s open-source Large Language architecture leveraging \\nLangChain revolutionizes the customer -company relationship \\nand CX. It enables companies to provide  efficient, \\n \\nFig. 2. User interface – Gradio framework'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 3, 'page_label': '4', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='Submitted to the 3rd International Conference on “Women in Science & Technology: Creating Sustainable Career”  \\n28 -30 December, 2023 \\npersonalized, and responsive support, ultimately leading  to \\ncustomer reten tion, increased customer value, and a more \\npositive brand image. As organizations continue to leverage the \\ncapabilities of LLMs, the landscape of customer service is \\nevolving into a more dynamic and customer-centric ecosystem. \\nThis paper dem onstrates a comparis on between various \\nmodel performances and evaluate s them on the basis of the \\nquality of response generated.  We ha ve compared \\nGOOGLE/FLAN-T5-XXL with GOOGLE/FLAN-T5-BASE, \\nand GOOGLE/FLAN-T5-SMALL and observed that the XXL \\nmodel outperforms the other  LLMs in the pro vided task. Each \\nmodel is posed with the same questions. \\nIn the future, Sahaay can access PDFs, Videos, Audio, and \\nother files to extract relevant informat ion about, for example,'),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 3, 'page_label': '4', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content=\"model is posed with the same questions. \\nIn the future, Sahaay can access PDFs, Videos, Audio, and \\nother files to extract relevant informat ion about, for example, \\nstudent activities, research work and innovation carried out by \\nBVM. This multimodal capability has th e potential to change \\nforever the way we interact with websites  and retrieve \\ninformation in much less time. \\nACKNOWLEDGMENT \\nThe authors would like to express their deepest appreciation \\nto the research facili ty provided at TRS BVM Laboratory for \\nencouraging multi-disciplinary collaborative research within \\nthe campus. We ’d also like to thank Birla Vishvakarma \\nMahavidyalaya (E ngineering College) for allowing us to \\nexperiment with the innovation on their website. \\nREFERENCES \\n[1] Asbjørn Følstad and Marita Skjuve. 201 9. Chatbots for customer \\nservice: user experience and motivatio n. In Proceedings of the 1st \\nInternational Conference on Conversational User Interfaces (CUI '19).\"),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 3, 'page_label': '4', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content=\"service: user experience and motivatio n. In Proceedings of the 1st \\nInternational Conference on Conversational User Interfaces (CUI '19). \\nAssociation for Computing Machinery, New York, NY, USA, Article 1, \\n1–9. https://doi.org/10.1145/3342775.3342784 \\n[2] Kim, S., Rawat, A. S., Zaheer, M., Jayasumana , S., Sadhanala, V., \\nJitkrittum, W., Menon, A. K., Fergus, R., & Kumar, S. (2023). \\nEmbedDistill: A Geometric Knowledge Distillation for Informa tion \\nRetrieval. ArXiv. /abs/2301.12005 \\n[3] Luiz Bonifacio, Hugo Abonizio, Marzieh Fadaee, and Rodrigo \\nNogueira. 20 22. InPars: Unsupervised Dataset Generation for \\nInformation Retrieval. In Proceedings of the 45th International ACM \\nSIGIR Conference on Research a nd Development in Information \\nRetrieval (SIGIR '22). Association for Computing Machinery, New \\nYork, NY, USA, 2387–2392. https://doi.org/10.1145/3477495.3531863 \\n[4] Su, Hongjin, Weijia Shi, Jungo Kasai, Yizhong Wang, Yushi Hu, Mari\"),\n",
       " Document(metadata={'producer': 'Microsoft® Word 2016', 'creator': 'Microsoft® Word 2016', 'creationdate': '2023-10-09T10:46:36+05:30', 'title': 'Paper Title (use style: paper title)', 'author': 'Keivalya Pandya;Dr Mehfuza Holia', 'moddate': '2023-10-09T10:46:36+05:30', 'source': '../data/pdf/2310.05421v1.pdf', 'total_pages': 4, 'page': 3, 'page_label': '4', 'source_file': '2310.05421v1.pdf', 'file_type': 'pdf'}, page_content='York, NY, USA, 2387–2392. https://doi.org/10.1145/3477495.3531863 \\n[4] Su, Hongjin, Weijia Shi, Jungo Kasai, Yizhong Wang, Yushi Hu, Mari \\nOstendorf, Wen Yih, Noah A. S mith, Luke Zettlemoyer, and Tao Yu. \\n\"One Embedder, Any Task: Instruction -Finetuned Text Embeddings.\" \\nArXiv, (2022). /abs/2212.09741. \\n[5] Johnson, Jeff, Matthijs Douze, and Hervé Jégou. \"Billion -scale \\nSimilarity Search with GPUs.\" ArXiv, (2017). Accessed Septem ber 28, \\n2023. /abs/1702.08734. \\n[6] Chung, Hyung W., Le Hou, Shayne Longpre , Barret Zoph, Yi Tay, \\nWilliam Fedus, Yu nxuan Li et al. \"Scaling Instruction -Finetuned \\nLanguage Models.\" ArXiv, (2022). Accessed September 28, 2023. \\n/abs/2210.11416 \\n[7] Abid, A., Abdalla, A. , Abid, A., Khan, D., Alfozan, A., & Zou, J. \\n(2019). Gradio: Hassle -Free Sharing and Testing of ML Models in the \\nWild. ArXiv. /abs/1906.02569')]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "### Text splitting get into chunks\n",
    "\n",
    "def split_documents(documents,chunk_size=1000,chunk_overlap=200):\n",
    "    \"\"\"Split documents into smaller chunks for better RAG performance\"\"\"\n",
    "    text_splitter = RecursiveCharacterTextSplitter(\n",
    "        chunk_size=chunk_size,\n",
    "        chunk_overlap=chunk_overlap,\n",
    "        length_function=len,\n",
    "        separators=[\"\\n\\n\", \"\\n\", \" \", \"\"]\n",
    "    )\n",
    "    split_docs = text_splitter.split_documents(documents)\n",
    "    print(f\"Split {len(documents)} documents into {len(split_docs)} chunks\")\n",
    "    \n",
    "    # Show example of a chunk\n",
    "    if split_docs:\n",
    "        print(f\"\\nExample chunk:\")\n",
    "        print(f\"Content: {split_docs[0].page_content[:200]}...\")\n",
    "        print(f\"Metadata: {split_docs[0].metadata}\")\n",
    "    \n",
    "    return split_docs\n",
    "chunks=split_documents(all_pdf_documents)\n",
    "chunks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "32d9efc6",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Loading embedding model: all-MiniLM-L6-v2\n",
      "Model loaded successfully. Embedding dimension: 384\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<__main__.EmbeddingManager at 0x13ea57ad0>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import numpy as np\n",
    "from sentence_transformers import SentenceTransformer\n",
    "import chromadb\n",
    "from chromadb.config import Settings\n",
    "import uuid\n",
    "from typing import List, Dict, Any, Tuple\n",
    "from sklearn.metrics.pairwise import cosine_similarity\n",
    "\n",
    "class EmbeddingManager:\n",
    "    \"\"\"Handles document embedding generation using SentenceTransformer\"\"\"\n",
    "    \n",
    "    def __init__(self, model_name: str = \"all-MiniLM-L6-v2\"):\n",
    "        \"\"\"\n",
    "        Initialize the embedding manager\n",
    "        \n",
    "        Args:\n",
    "            model_name: HuggingFace model name for sentence embeddings\n",
    "        \"\"\"\n",
    "        self.model_name = model_name\n",
    "        self.model = None\n",
    "        self._load_model()\n",
    "\n",
    "    def _load_model(self):\n",
    "        \"\"\"Load the SentenceTransformer model\"\"\"\n",
    "        try:\n",
    "            print(f\"Loading embedding model: {self.model_name}\")\n",
    "            self.model = SentenceTransformer(self.model_name)\n",
    "            print(f\"Model loaded successfully. Embedding dimension: {self.model.get_sentence_embedding_dimension()}\")\n",
    "        except Exception as e:\n",
    "            print(f\"Error loading model {self.model_name}: {e}\")\n",
    "            raise\n",
    "\n",
    "    def generate_embeddings(self, texts: List[str]) -> np.ndarray:\n",
    "        \"\"\"\n",
    "        Generate embeddings for a list of texts\n",
    "        \n",
    "        Args:\n",
    "            texts: List of text strings to embed\n",
    "            \n",
    "        Returns:\n",
    "            numpy array of embeddings with shape (len(texts), embedding_dim)\n",
    "        \"\"\"\n",
    "        if not self.model:\n",
    "            raise ValueError(\"Model not loaded\")\n",
    "        \n",
    "        print(f\"Generating embeddings for {len(texts)} texts...\")\n",
    "        embeddings = self.model.encode(texts, show_progress_bar=True)\n",
    "        print(f\"Generated embeddings with shape: {embeddings.shape}\")\n",
    "        return embeddings\n",
    "\n",
    "\n",
    "## initialize the embedding manager\n",
    "\n",
    "embedding_manager=EmbeddingManager()\n",
    "embedding_manager"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "9f36ff98",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Vector store initialized. Collection: pdf_documents\n",
      "Existing documents in collection: 124\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<__main__.VectorStore at 0x13e7c7350>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class VectorStore:\n",
    "    \"\"\"Manages document embeddings in a ChromaDB vector store\"\"\"\n",
    "    \n",
    "    def __init__(self, collection_name: str = \"pdf_documents\", persist_directory: str = \"../data/vector_store\"):\n",
    "        \"\"\"\n",
    "        Initialize the vector store\n",
    "        \n",
    "        Args:\n",
    "            collection_name: Name of the ChromaDB collection\n",
    "            persist_directory: Directory to persist the vector store\n",
    "        \"\"\"\n",
    "        self.collection_name = collection_name\n",
    "        self.persist_directory = persist_directory\n",
    "        self.client = None\n",
    "        self.collection = None\n",
    "        self._initialize_store()\n",
    "\n",
    "    def _initialize_store(self):\n",
    "        \"\"\"Initialize ChromaDB client and collection\"\"\"\n",
    "        try:\n",
    "            # Create persistent ChromaDB client\n",
    "            os.makedirs(self.persist_directory, exist_ok=True)\n",
    "            self.client = chromadb.PersistentClient(path=self.persist_directory)\n",
    "            \n",
    "            # Get or create collection\n",
    "            self.collection = self.client.get_or_create_collection(\n",
    "                name=self.collection_name,\n",
    "                metadata={\"description\": \"PDF document embeddings for RAG\"}\n",
    "            )\n",
    "            print(f\"Vector store initialized. Collection: {self.collection_name}\")\n",
    "            print(f\"Existing documents in collection: {self.collection.count()}\")\n",
    "            \n",
    "        except Exception as e:\n",
    "            print(f\"Error initializing vector store: {e}\")\n",
    "            raise\n",
    "\n",
    "    def add_documents(self, documents: List[Any], embeddings: np.ndarray):\n",
    "        \"\"\"\n",
    "        Add documents and their embeddings to the vector store\n",
    "        \n",
    "        Args:\n",
    "            documents: List of LangChain documents\n",
    "            embeddings: Corresponding embeddings for the documents\n",
    "        \"\"\"\n",
    "        if len(documents) != len(embeddings):\n",
    "            raise ValueError(\"Number of documents must match number of embeddings\")\n",
    "        \n",
    "        print(f\"Adding {len(documents)} documents to vector store...\")\n",
    "        \n",
    "        # Prepare data for ChromaDB\n",
    "        ids = []\n",
    "        metadatas = []\n",
    "        documents_text = []\n",
    "        embeddings_list = []\n",
    "        \n",
    "        for i, (doc, embedding) in enumerate(zip(documents, embeddings)):\n",
    "            # Generate unique ID\n",
    "            doc_id = f\"doc_{uuid.uuid4().hex[:8]}_{i}\"\n",
    "            ids.append(doc_id)\n",
    "            \n",
    "            # Prepare metadata\n",
    "            metadata = dict(doc.metadata)\n",
    "            metadata['doc_index'] = i\n",
    "            metadata['content_length'] = len(doc.page_content)\n",
    "            metadatas.append(metadata)\n",
    "            # Document content\n",
    "            documents_text.append(doc.page_content)\n",
    "            \n",
    "            # Embedding\n",
    "            embeddings_list.append(embedding.tolist())\n",
    "        \n",
    "        # Add to collection\n",
    "        try:\n",
    "            self.collection.add(\n",
    "                ids=ids,\n",
    "                embeddings=embeddings_list,\n",
    "                metadatas=metadatas,\n",
    "                documents=documents_text\n",
    "            )\n",
    "            print(f\"Successfully added {len(documents)} documents to vector store\")\n",
    "            print(f\"Total documents in collection: {self.collection.count()}\")\n",
    "            \n",
    "        except Exception as e:\n",
    "            print(f\"Error adding documents to vector store: {e}\")\n",
    "            raise\n",
    "\n",
    "vectorstore=VectorStore()\n",
    "vectorstore"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "a4c2e333",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generating embeddings for 62 texts...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Batches: 100%|██████████| 2/2 [00:01<00:00,  1.54it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generated embeddings with shape: (62, 384)\n",
      "Adding 62 documents to vector store...\n",
      "Successfully added 62 documents to vector store\n",
      "Total documents in collection: 186\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "### Convert the text to embeddings\n",
    "texts=[doc.page_content for doc in chunks]\n",
    "\n",
    "## Generate the Embeddings\n",
    "\n",
    "embeddings=embedding_manager.generate_embeddings(texts)\n",
    "\n",
    "##store int he vector dtaabase\n",
    "vectorstore.add_documents(chunks,embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "4c234a53",
   "metadata": {},
   "outputs": [],
   "source": [
    "class RAGRetriever:\n",
    "    \"\"\"Handles query-based retrieval from the vector store\"\"\"\n",
    "    \n",
    "    def __init__(self, vector_store: VectorStore, embedding_manager: EmbeddingManager):\n",
    "        \"\"\"\n",
    "        Initialize the retriever\n",
    "        \n",
    "        Args:\n",
    "            vector_store: Vector store containing document embeddings\n",
    "            embedding_manager: Manager for generating query embeddings\n",
    "        \"\"\"\n",
    "        self.vector_store = vector_store\n",
    "        self.embedding_manager = embedding_manager\n",
    "\n",
    "    def retrieve(self, query: str, top_k: int = 5, score_threshold: float = 0.0) -> List[Dict[str, Any]]:\n",
    "        \"\"\"\n",
    "        Retrieve relevant documents for a query\n",
    "        \n",
    "        Args:\n",
    "            query: The search query\n",
    "            top_k: Number of top results to return\n",
    "            score_threshold: Minimum similarity score threshold\n",
    "            \n",
    "        Returns:\n",
    "            List of dictionaries containing retrieved documents and metadata\n",
    "        \"\"\"\n",
    "        print(f\"Retrieving documents for query: '{query}'\")\n",
    "        print(f\"Top K: {top_k}, Score threshold: {score_threshold}\")\n",
    "        \n",
    "        # Generate query embedding\n",
    "        query_embedding = self.embedding_manager.generate_embeddings([query])[0]\n",
    "        \n",
    "        # Search in vector store\n",
    "        try:\n",
    "            results = self.vector_store.collection.query(\n",
    "                query_embeddings=[query_embedding.tolist()],\n",
    "                n_results=top_k\n",
    "            )\n",
    "            \n",
    "            # Process results\n",
    "            retrieved_docs = []\n",
    "            \n",
    "            if results['documents'] and results['documents'][0]:\n",
    "                documents = results['documents'][0]\n",
    "                metadatas = results['metadatas'][0]\n",
    "                distances = results['distances'][0]\n",
    "                ids = results['ids'][0]\n",
    "                \n",
    "                for i, (doc_id, document, metadata, distance) in enumerate(zip(ids, documents, metadatas, distances)):\n",
    "                    # Convert distance to similarity score (ChromaDB uses cosine distance)\n",
    "                    similarity_score = 1 - distance\n",
    "                    \n",
    "                    if similarity_score >= score_threshold:\n",
    "                        retrieved_docs.append({\n",
    "                            'id': doc_id,\n",
    "                            'content': document,\n",
    "                            'metadata': metadata,\n",
    "                            'similarity_score': similarity_score,\n",
    "                            'distance': distance,\n",
    "                            'rank': i + 1\n",
    "                        })\n",
    "                \n",
    "                print(f\"Retrieved {len(retrieved_docs)} documents (after filtering)\")\n",
    "            else:\n",
    "                print(\"No documents found\")\n",
    "            \n",
    "            return retrieved_docs\n",
    "            \n",
    "        except Exception as e:\n",
    "            print(f\"Error during retrieval: {e}\")\n",
    "            return []\n",
    "\n",
    "rag_retriever=RAGRetriever(vectorstore,embedding_manager)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "402e545f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Retrieving documents for query: 'What is attention is all you need'\n",
      "Top K: 5, Score threshold: 0.0\n",
      "Generating embeddings for 1 texts...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Batches: 100%|██████████| 1/1 [00:00<00:00,  1.04it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generated embeddings with shape: (1, 384)\n",
      "Retrieved 0 documents (after filtering)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rag_retriever.retrieve(\"What is attention is all you need\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "51204a01",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f04990b7",
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "8e8d1c27",
   "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": "3.12.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}