notebooks/head_view_xlm.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from bertviz import head_view\n",
    "from transformers import XLMTokenizer, XLMModel"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "model_version = 'xlm-mlm-ende-1024'\n",
    "model = XLMModel.from_pretrained(model_version, output_attentions=True)\n",
    "tokenizer = XLMTokenizer.from_pretrained(model_version)\n",
    "\n",
    "text = \"Do you speak German? Ja, ich spreche Deutsch.\"\n",
    "inputs = tokenizer.encode_plus(text, return_tensors='pt', add_special_tokens=True)\n",
    "input_ids = inputs['input_ids']\n",
    "attention = model(input_ids)[-1]\n",
    "input_id_list = input_ids[0].tolist() # Batch index 0\n",
    "tokens = tokenizer.convert_ids_to_tokens(input_id_list)\n",
    "head_view(attention, tokens)"
   ]
  }
 ],
 "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"
  },
  "pycharm": {
   "stem_cell": {
    "cell_type": "raw",
    "metadata": {
     "collapsed": false
    },
    "source": []
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}