notebooks/neuron_view_roberta.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Neuron View\n",
    "The attention-head view visualizes attention, as well as query and key values, in a particuler attention head."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from bertviz.transformers_neuron_view import RobertaModel, RobertaTokenizer\n",
    "from bertviz.neuron_view import show"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Usage\n",
    "* **Hover** over any of the tokens on the left side of the visualization to filter attention from that token.\n",
    "* Then **click** on the **plus** icon that is revealed when hovering. This shows the query vectors, key vectors, and intermediate computations for the attention weights (<span style='color:blue'>blue</span>=positive, <span style='color:#ff6318'>orange</span>=negative).  \n",
    "* Once in the expanded view, **hover** over any other **token** on the left to see the associated attention computations.\n",
    "* **Click** on the **Layer** or **Head** drop-downs to change the model layer or head (zero-indexed)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_type = 'roberta'\n",
    "model_version = 'roberta-base'\n",
    "model = RobertaModel.from_pretrained(model_version)\n",
    "tokenizer = RobertaTokenizer.from_pretrained(model_version)\n",
    "sentence_a = \"The cat sat on the mat\"\n",
    "sentence_b = \"The cat lay on the rug\"\n",
    "show(model, model_type, tokenizer, sentence_a, sentence_b)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}