Upload scandl2_demo.ipynb (#2)

- Upload scandl2_demo.ipynb (14a64a0105634aaa5b22dc7359c77426273532db)


Co-authored-by: Le Minh Truong <minhgou@users.noreply.huggingface.co>

scandl2_demo.ipynb

@@ -0,0 +1,943 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a8049e2d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "%pip uninstall -y gradio gradio-client\n",
+    "%pip install \"huggingface-hub>=0.16.4,<0.18\"\n",
+    "%pip check"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "1d32e5f6",
+   "metadata": {},
+   "source": [
+    "# ScanDL2 Demo"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 107,
+   "id": "73158dd6",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from ScanDL2 import ScanDL2\n",
+    "import torch\n",
+    "import sys\n",
+    "import torch\n",
+    "from unittest.mock import patch"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 110,
+   "id": "6cb55d37",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "sample_sentences = [\n",
+    "    \"The quick brown fox jumps over the lazy dog.\",\n",
+    "]\n",
+    "sample_paragraphs = [\n",
+    "    \"The sun was shining brightly in the clear blue sky. Birds chirped melodiously from the trees, filling the air with a cheerful tune. It was the perfect day for a walk in the park.\",\n",
+    "    \"She had always loved the scent of fresh coffee in the morning. As she sipped her espresso, she glanced at the newspaper headlines. Another day, another adventure awaited her.\",\n",
+    "    \"The conference room was filled with eager attendees. The keynote speaker took the stage, adjusting the microphone. With a confident smile, he began his presentation on the future of artificial intelligence.\",\n",
+    "    \"They packed their bags and set off on their long-awaited road trip. The open road stretched ahead, inviting them to explore new places. With music playing softly in the background, they felt a sense of freedom like never before.\",\n",
+    "]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "762cf32a",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def display_scandl_scanpath(text, output, sample_index=0):\n",
+    "    words = output[\"predicted_sp_words\"][sample_index]\n",
+    "    positions = output[\"predicted_sp_ids\"][sample_index]\n",
+    "    durations = output[\"predicted_fix_durs\"][sample_index]\n",
+    "\n",
+    "    if not (len(words) == len(positions) == len(durations)):\n",
+    "        raise ValueError(\"Word, position, and duration lengths do not match.\")\n",
+    "\n",
+    "    rows = []\n",
+    "\n",
+    "    for step in range(len(positions)):\n",
+    "        word = str(words[step]).replace(\"|\", r\"\\|\")\n",
+    "        rows.append(\n",
+    "            f\"| {step + 1} | {word} | {positions[step]} \"\n",
+    "            f\"| {float(durations[step]):.1f} |\"\n",
+    "        )\n",
+    "\n",
+    "    display(Markdown(\n",
+    "        f\"**Input:** {text}\\n\\n\"\n",
+    "        f\"**Scanpath:** {positions}\\n\\n\"\n",
+    "        f\"**Fixated sequence:** {' → '.join(words)}\\n\\n\"\n",
+    "        \"| Fixation | Word | Position | Duration (ms) |\\n\"\n",
+    "        \"|---|---|---|---|\\n\"\n",
+    "        + \"\\n\".join(rows)\n",
+    "    ))\n",
+    "\n",
+    "def run_scandl_raw_text_input(model, text):\n",
+    "    torch.manual_seed(0)\n",
+    "\n",
+    "    with torch.no_grad():\n",
+    "        return model(texts=[text])"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "34e2e22e",
+   "metadata": {},
+   "source": [
+    "## ScanDL2 — Sentences"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 111,
+   "id": "9be2fa4a",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Logging to /var/folders/vg/2mjpcbv55m966znhbf9c5k580000gn/T/openai-2026-10-27-16-28-41-658594\n",
+      "### Loading ScanDL Diffusion Module ...\n",
+      "### Total number of parameters: 130855296\n",
+      "\t\t### ScanDL Module generates fixation locations ...\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "1it [01:08, 68.80s/it]\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\t\t### FixDur Module generates fixation durations ...\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "1it [00:00, 12.33it/s]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "fixdur original sn: ['The', 'quick', 'brown', 'fox', 'jumps', 'over', 'the', 'lazy', 'dog.']\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n"
+     ]
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "### Sentence 1"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "**Input:** The quick brown fox jumps over the lazy dog.\n",
+       "\n",
+       "**Scanpath:** [1, 2, 3, 4, 5, 5, 6, 8, 8]\n",
+       "\n",
+       "**Fixated sequence:** The → quick → brown → fox → jumps → jumps → over → lazy → lazy\n",
+       "\n",
+       "| Fixation | Word | Position | Duration (ms) |\n",
+       "|---|---|---|---|\n",
+       "| 1 | The | 1 | 191.8 |\n",
+       "| 2 | quick | 2 | 188.5 |\n",
+       "| 3 | brown | 3 | 201.7 |\n",
+       "| 4 | fox | 4 | 207.9 |\n",
+       "| 5 | jumps | 5 | 219.3 |\n",
+       "| 6 | jumps | 5 | 229.0 |\n",
+       "| 7 | over | 6 | 235.7 |\n",
+       "| 8 | lazy | 8 | 221.9 |\n",
+       "| 9 | lazy | 8 | 173.3 |"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "with patch.object(sys, \"argv\", [sys.argv[0]]):\n",
+    "    scandl2_sent = ScanDL2(\n",
+    "    text_type=\"sentence\",\n",
+    "    bsz=2,\n",
+    "    save=None,\n",
+    "    filename=None,\n",
+    ")\n",
+    "scandl2_sent.eval()\n",
+    "\n",
+    "with torch.no_grad():\n",
+    "    sentence_output = scandl2_sent(sample_sentences)\n",
+    "\n",
+    "for i, text in enumerate(sample_sentences):\n",
+    "    display(Markdown(f\"### Sentence {i + 1}\"))\n",
+    "    display_scandl_scanpath(text, sentence_output, sample_index=i)\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "2887484f",
+   "metadata": {},
+   "source": [
+    "## ScanDL2 — Paragraphs"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 109,
+   "id": "6e695749",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Logging to /var/folders/vg/2mjpcbv55m966znhbf9c5k580000gn/T/openai-2026-10-27-16-08-03-314381\n",
+      "### Loading ScanDL Diffusion Module ...\n",
+      "### Total number of parameters: 131630560\n",
+      "\t\t### ScanDL Module generates fixation locations ...\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "2it [11:09, 334.98s/it]\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "\t\t### FixDur Module generates fixation durations ...\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "1it [00:00,  2.03it/s]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "fixdur original sn: ['The', 'sun', 'was', 'shining', 'brightly', 'in', 'the', 'clear', 'blue', 'sky.', 'Birds', 'chirped', 'melodiously', 'from', 'the', 'trees,', 'filling', 'the', 'air', 'with', 'a', 'cheerful', 'tune.', 'It', 'was', 'the', 'perfect', 'day', 'for', 'a', 'walk', 'in', 'the', 'park.']\n",
+      "fixdur original sn: ['She', 'had', 'always', 'loved', 'the', 'scent', 'of', 'fresh', 'coffee', 'in', 'the', 'morning.', 'As', 'she', 'sipped', 'her', 'espresso,', 'she', 'glanced', 'at', 'the', 'newspaper', 'headlines.', 'Another', 'day,', 'another', 'adventure', 'awaited', 'her.']\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "2it [00:00,  2.36it/s]"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "fixdur original sn: ['The', 'conference', 'room', 'was', 'filled', 'with', 'eager', 'attendees.', 'The', 'keynote', 'speaker', 'took', 'the', 'stage,', 'adjusting', 'the', 'microphone.', 'With', 'a', 'confident', 'smile,', 'he', 'began', 'his', 'presentation', 'on', 'the', 'future', 'of', 'artificial', 'intelligence.']\n",
+      "fixdur original sn: ['They', 'packed', 'their', 'bags', 'and', 'set', 'off', 'on', 'their', 'long-awaited', 'road', 'trip.', 'The', 'open', 'road', 'stretched', 'ahead,', 'inviting', 'them', 'to', 'explore', 'new', 'places.', 'With', 'music', 'playing', 'softly', 'in', 'the', 'background,', 'they', 'felt', 'a', 'sense', 'of', 'freedom', 'like', 'never', 'before.']\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\n"
+     ]
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "### Paragraph 1"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "**Input:** The sun was shining brightly in the clear blue sky. Birds chirped melodiously from the trees, filling the air with a cheerful tune. It was the perfect day for a walk in the park.\n",
+       "\n",
+       "**Scanpath:** [2, 3, 4, 4, 5, 8, 7, 9, 11, 11, 13, 12, 12, 13, 14, 17, 17, 16, 18, 20, 22, 22, 23, 25, 27, 29, 28, 31, 31, 32, 34, 34]\n",
+       "\n",
+       "**Fixated sequence:** sun → was → shining → shining → brightly → clear → the → blue → Birds → Birds → melodiously → chirped → chirped → melodiously → from → filling → filling → trees, → the → with → cheerful → cheerful → tune. → was → perfect → for → day → walk → walk → in → park. → park.\n",
+       "\n",
+       "| Fixation | Word | Position | Duration (ms) |\n",
+       "|---|---|---|---|\n",
+       "| 1 | sun | 2 | 221.6 |\n",
+       "| 2 | was | 3 | 197.2 |\n",
+       "| 3 | shining | 4 | 212.7 |\n",
+       "| 4 | shining | 4 | 220.1 |\n",
+       "| 5 | brightly | 5 | 228.2 |\n",
+       "| 6 | clear | 8 | 226.2 |\n",
+       "| 7 | the | 7 | 224.6 |\n",
+       "| 8 | blue | 9 | 220.1 |\n",
+       "| 9 | Birds | 11 | 219.7 |\n",
+       "| 10 | Birds | 11 | 221.9 |\n",
+       "| 11 | melodiously | 13 | 227.9 |\n",
+       "| 12 | chirped | 12 | 231.2 |\n",
+       "| 13 | chirped | 12 | 221.8 |\n",
+       "| 14 | melodiously | 13 | 218.9 |\n",
+       "| 15 | from | 14 | 221.9 |\n",
+       "| 16 | filling | 17 | 230.9 |\n",
+       "| 17 | filling | 17 | 210.3 |\n",
+       "| 18 | trees, | 16 | 223.7 |\n",
+       "| 19 | the | 18 | 222.9 |\n",
+       "| 20 | with | 20 | 228.0 |\n",
+       "| 21 | cheerful | 22 | 223.6 |\n",
+       "| 22 | cheerful | 22 | 219.1 |\n",
+       "| 23 | tune. | 23 | 212.6 |\n",
+       "| 24 | was | 25 | 209.8 |\n",
+       "| 25 | perfect | 27 | 206.0 |\n",
+       "| 26 | for | 29 | 224.1 |\n",
+       "| 27 | day | 28 | 210.8 |\n",
+       "| 28 | walk | 31 | 196.1 |\n",
+       "| 29 | walk | 31 | 193.6 |\n",
+       "| 30 | in | 32 | 183.9 |\n",
+       "| 31 | park. | 34 | 189.7 |\n",
+       "| 32 | park. | 34 | 187.6 |"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "### Paragraph 2"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "**Input:** She had always loved the scent of fresh coffee in the morning. As she sipped her espresso, she glanced at the newspaper headlines. Another day, another adventure awaited her.\n",
+       "\n",
+       "**Scanpath:** [1, 2, 3, 4, 6, 8, 9, 11, 12, 15, 15, 15, 17, 19, 23, 24, 25, 27, 28, 28, 29]\n",
+       "\n",
+       "**Fixated sequence:** She → had → always → loved → scent → fresh → coffee → the → morning. → sipped → sipped → sipped → espresso, → glanced → headlines. → Another → day, → adventure → awaited → awaited → her.\n",
+       "\n",
+       "| Fixation | Word | Position | Duration (ms) |\n",
+       "|---|---|---|---|\n",
+       "| 1 | She | 1 | 220.6 |\n",
+       "| 2 | had | 2 | 195.1 |\n",
+       "| 3 | always | 3 | 211.5 |\n",
+       "| 4 | loved | 4 | 218.4 |\n",
+       "| 5 | scent | 6 | 226.8 |\n",
+       "| 6 | fresh | 8 | 226.2 |\n",
+       "| 7 | coffee | 9 | 223.3 |\n",
+       "| 8 | the | 11 | 219.0 |\n",
+       "| 9 | morning. | 12 | 217.6 |\n",
+       "| 10 | sipped | 15 | 219.6 |\n",
+       "| 11 | sipped | 15 | 225.9 |\n",
+       "| 12 | sipped | 15 | 227.9 |\n",
+       "| 13 | espresso, | 17 | 220.1 |\n",
+       "| 14 | glanced | 19 | 216.8 |\n",
+       "| 15 | headlines. | 23 | 219.4 |\n",
+       "| 16 | Another | 24 | 228.6 |\n",
+       "| 17 | day, | 25 | 207.4 |\n",
+       "| 18 | adventure | 27 | 222.9 |\n",
+       "| 19 | awaited | 28 | 220.6 |\n",
+       "| 20 | awaited | 28 | 227.6 |\n",
+       "| 21 | her. | 29 | 222.0 |"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "### Paragraph 3"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "**Input:** The conference room was filled with eager attendees. The keynote speaker took the stage, adjusting the microphone. With a confident smile, he began his presentation on the future of artificial intelligence.\n",
+       "\n",
+       "**Scanpath:** [1, 2, 3, 3, 5, 5, 7, 7, 8, 9, 10, 10, 12, 11, 14, 15, 17, 19, 21, 25, 23, 27, 28, 30, 31]\n",
+       "\n",
+       "**Fixated sequence:** The → conference → room → room → filled → filled → eager → eager → attendees. → The → keynote → keynote → took → speaker → stage, → adjusting → microphone. → a → smile, → presentation → began → the → future → artificial → intelligence.\n",
+       "\n",
+       "| Fixation | Word | Position | Duration (ms) |\n",
+       "|---|---|---|---|\n",
+       "| 1 | The | 1 | 220.4 |\n",
+       "| 2 | conference | 2 | 195.4 |\n",
+       "| 3 | room | 3 | 211.5 |\n",
+       "| 4 | room | 3 | 218.5 |\n",
+       "| 5 | filled | 5 | 226.9 |\n",
+       "| 6 | filled | 5 | 225.9 |\n",
+       "| 7 | eager | 7 | 223.4 |\n",
+       "| 8 | eager | 7 | 219.0 |\n",
+       "| 9 | attendees. | 8 | 217.8 |\n",
+       "| 10 | The | 9 | 219.9 |\n",
+       "| 11 | keynote | 10 | 226.0 |\n",
+       "| 12 | keynote | 10 | 228.4 |\n",
+       "| 13 | took | 12 | 220.2 |\n",
+       "| 14 | speaker | 11 | 217.0 |\n",
+       "| 15 | stage, | 14 | 219.7 |\n",
+       "| 16 | adjusting | 15 | 228.9 |\n",
+       "| 17 | microphone. | 17 | 207.9 |\n",
+       "| 18 | a | 19 | 222.7 |\n",
+       "| 19 | smile, | 21 | 220.9 |\n",
+       "| 20 | presentation | 25 | 227.4 |\n",
+       "| 21 | began | 23 | 222.1 |\n",
+       "| 22 | the | 27 | 218.0 |\n",
+       "| 23 | future | 28 | 210.6 |\n",
+       "| 24 | artificial | 30 | 208.1 |\n",
+       "| 25 | intelligence. | 31 | 203.7 |"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "### Paragraph 4"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "**Input:** They packed their bags and set off on their long-awaited road trip. The open road stretched ahead, inviting them to explore new places. With music playing softly in the background, they felt a sense of freedom like never before.\n",
+       "\n",
+       "**Scanpath:** [1, 2, 3, 4, 6, 7, 8, 9, 10, 11, 12, 13, 14, 16, 16, 18, 17, 19, 19, 21, 21, 22, 22, 23, 25, 25, 27, 29, 30, 31, 32, 33, 36, 37, 36, 39]\n",
+       "\n",
+       "**Fixated sequence:** They → packed → their → bags → set → off → on → their → long-awaited → road → trip. → The → open → stretched → stretched → inviting → ahead, → them → them → explore → explore → new → new → places. → music → music → softly → the → background, → they → felt → a → freedom → like → freedom → before.\n",
+       "\n",
+       "| Fixation | Word | Position | Duration (ms) |\n",
+       "|---|---|---|---|\n",
+       "| 1 | They | 1 | 222.5 |\n",
+       "| 2 | packed | 2 | 198.3 |\n",
+       "| 3 | their | 3 | 213.7 |\n",
+       "| 4 | bags | 4 | 221.3 |\n",
+       "| 5 | set | 6 | 229.2 |\n",
+       "| 6 | off | 7 | 226.6 |\n",
+       "| 7 | on | 8 | 225.4 |\n",
+       "| 8 | their | 9 | 220.7 |\n",
+       "| 9 | long-awaited | 10 | 221.2 |\n",
+       "| 10 | road | 11 | 223.3 |\n",
+       "| 11 | trip. | 12 | 229.2 |\n",
+       "| 12 | The | 13 | 232.8 |\n",
+       "| 13 | open | 14 | 223.0 |\n",
+       "| 14 | stretched | 16 | 220.0 |\n",
+       "| 15 | stretched | 16 | 223.2 |\n",
+       "| 16 | inviting | 18 | 232.0 |\n",
+       "| 17 | ahead, | 17 | 211.8 |\n",
+       "| 18 | them | 19 | 224.5 |\n",
+       "| 19 | them | 19 | 224.2 |\n",
+       "| 20 | explore | 21 | 228.6 |\n",
+       "| 21 | explore | 21 | 224.7 |\n",
+       "| 22 | new | 22 | 220.0 |\n",
+       "| 23 | new | 22 | 214.0 |\n",
+       "| 24 | places. | 23 | 210.9 |\n",
+       "| 25 | music | 25 | 207.4 |\n",
+       "| 26 | music | 25 | 224.4 |\n",
+       "| 27 | softly | 27 | 212.3 |\n",
+       "| 28 | the | 29 | 198.9 |\n",
+       "| 29 | background, | 30 | 196.8 |\n",
+       "| 30 | they | 31 | 187.3 |\n",
+       "| 31 | felt | 32 | 192.9 |\n",
+       "| 32 | a | 33 | 191.2 |\n",
+       "| 33 | freedom | 36 | 174.0 |\n",
+       "| 34 | like | 37 | 186.6 |\n",
+       "| 35 | freedom | 36 | 182.3 |\n",
+       "| 36 | before. | 39 | 168.4 |"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "with patch.object(sys, \"argv\", [sys.argv[0]]):\n",
+    "    scandl2_par = ScanDL2(\n",
+    "    text_type=\"paragraph\",\n",
+    "    bsz=2,\n",
+    "    save=None,\n",
+    "    filename=None,\n",
+    ")\n",
+    "scandl2_par.eval()\n",
+    "\n",
+    "with torch.no_grad():\n",
+    "    paragraph_output = scandl2_par(sample_paragraphs)\n",
+    "\n",
+    "for i, text in enumerate(sample_paragraphs):\n",
+    "    display(Markdown(f\"### Paragraph {i + 1}\"))\n",
+    "    display_scandl_scanpath(text, paragraph_output, sample_index=i)\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "1c44a7f5",
+   "metadata": {},
+   "source": [
+    "# Eyettention Demo"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "c33d6269",
+   "metadata": {},
+   "source": [
+    "### Chinese Language / BSC Dataset"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "14a3e7ef",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from IPython.display import display, Markdown\n",
+    "import warnings\n",
+    "\n",
+    "warnings.filterwarnings(\"ignore\")\n",
+    "\n",
+    "def display_scanpath(text, scanpath, dataset=\"BSC\", previous_scanpath=None):\n",
+    "    if dataset == \"BSC\":\n",
+    "        units = [char for char in text if not char.isspace()]\n",
+    "    elif dataset == \"celer\":\n",
+    "        units = text.split()\n",
+    "    else:\n",
+    "        raise ValueError(\"dataset must be 'BSC' or 'celer'\")\n",
+    "\n",
+    "    positions = scanpath[0].tolist()\n",
+    "\n",
+    "    # Match the model's handling of an omitted start token.\n",
+    "    if previous_scanpath is not None:\n",
+    "        previous_scanpath = list(previous_scanpath)\n",
+    "        if not previous_scanpath or previous_scanpath[0] != 0:\n",
+    "            previous_scanpath = [0] + previous_scanpath\n",
+    "\n",
+    "    rows = []\n",
+    "    sequence = []\n",
+    "\n",
+    "    for step, position in enumerate(positions):\n",
+    "        if position == len(units) + 1:  # End token\n",
+    "            break\n",
+    "        if position == 0:             # Start token\n",
+    "            continue\n",
+    "\n",
+    "        unit = (\n",
+    "            units[position - 1]\n",
+    "            if 1 <= position <= len(units)\n",
+    "            else \"?\"\n",
+    "        )\n",
+    "\n",
+    "        source = (\n",
+    "            \"Observed\"\n",
+    "            if previous_scanpath is not None and step < len(previous_scanpath)\n",
+    "            else \"Generated\"\n",
+    "        )\n",
+    "\n",
+    "        sequence.append(unit)\n",
+    "        rows.append(\n",
+    "            f\"| {len(rows) + 1} | {unit} | {position} | {source} |\"\n",
+    "        )\n",
+    "\n",
+    "    display(Markdown(\n",
+    "        f\"**Input:** {text}\\n\\n\"\n",
+    "        f\"**Scanpath:** {scanpath.tolist()[0]}\\n\\n\"\n",
+    "        f\"**Fixated sequence:** {' → '.join(sequence)}\\n\\n\"\n",
+    "        \"| Fixation | Word / Character | Position | Source |\\n\"\n",
+    "        \"|---|---|---|---|\\n\"\n",
+    "        + \"\\n\".join(rows)\n",
+    "    ))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "23163c57",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "keeping Bert with pre-trained weights\n"
+     ]
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "## Chinese Raw-text generation"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "**Input:** 中国选手在男子滑雪比赛中有望蝉联冠军\n",
+       "\n",
+       "**Scanpath:** [0, 1, 1, 4, 3]\n",
+       "\n",
+       "**Fixated sequence:** 中 → 中 → 手 → 选\n",
+       "\n",
+       "| Fixation | Word / Character | Position | Source |\n",
+       "|---|---|---|---|\n",
+       "| 1 | 中 | 1 | Generated |\n",
+       "| 2 | 中 | 1 | Generated |\n",
+       "| 3 | 手 | 4 | Generated |\n",
+       "| 4 | 选 | 3 | Generated |"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "---\n",
+       "\n",
+       "## Chinese Continuation from a previous scanpath"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "**Input:** 中国选手在男子滑雪比赛中有望蝉联冠军\n",
+       "\n",
+       "**Scanpath:** [0, 1, 2, 3, 4, 4, 7, 9, 11]\n",
+       "\n",
+       "**Fixated sequence:** 中 → 国 → 选 → 手 → 手 → 子 → 雪 → 赛\n",
+       "\n",
+       "| Fixation | Word / Character | Position | Source |\n",
+       "|---|---|---|---|\n",
+       "| 1 | 中 | 1 | Observed |\n",
+       "| 2 | 国 | 2 | Observed |\n",
+       "| 3 | 选 | 3 | Observed |\n",
+       "| 4 | 手 | 4 | Generated |\n",
+       "| 5 | 手 | 4 | Generated |\n",
+       "| 6 | 子 | 7 | Generated |\n",
+       "| 7 | 雪 | 9 | Generated |\n",
+       "| 8 | 赛 | 11 | Generated |"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "from Eyettention import EyettentionRawTextInference\n",
+    "\n",
+    "chinese_text = \"中国选手在男子滑雪比赛中有望蝉联冠军\"\n",
+    "chinese_previous_scanpath = [1,2,3]\n",
+    "\n",
+    "chinese_model = EyettentionRawTextInference(\n",
+    "    checkpoint_path=\"Eyettention/results/BSC/Eyettention_chinese.pth\",\n",
+    "    dataset=\"BSC\",\n",
+    "    device=\"cpu\",\n",
+    ")\n",
+    "\n",
+    "def run_chinese_raw_text_input(model, text, max_pred_len=5):\n",
+    "    torch.manual_seed(42)\n",
+    "\n",
+    "    scanpath, _ = model.generate_from_chinese_text(\n",
+    "        text=text,\n",
+    "        max_pred_len=max_pred_len,\n",
+    "        previous_scanpath=None\n",
+    "    )\n",
+    "    \n",
+    "    return scanpath\n",
+    "\n",
+    "def run_chinese_previous_scanpath(model, text, max_pred_len=5, previous_scanpath=None):\n",
+    "    torch.manual_seed(42)\n",
+    "\n",
+    "    scanpath, _ = model.generate_from_chinese_text(\n",
+    "        text=text,\n",
+    "        max_pred_len=max_pred_len,\n",
+    "        previous_scanpath= previous_scanpath\n",
+    "    )\n",
+    "\n",
+    "    return scanpath\n",
+    "\n",
+    "\n",
+    "display(Markdown(\"## Chinese Raw-text generation\"))\n",
+    "\n",
+    "raw_text = run_chinese_raw_text_input(chinese_model, chinese_text)\n",
+    "display_scanpath(chinese_text, scanpath=raw_text, dataset=\"BSC\")\n",
+    "\n",
+    "\n",
+    "display(Markdown(\"---\\n\\n## Chinese Continuation from a previous scanpath\"))\n",
+    "\n",
+    "\n",
+    "continuation = run_chinese_previous_scanpath(\n",
+    "    chinese_model,\n",
+    "    chinese_text,\n",
+    "    previous_scanpath=chinese_previous_scanpath,\n",
+    ")\n",
+    "\n",
+    "display_scanpath(\n",
+    "    chinese_text,\n",
+    "    scanpath=continuation,\n",
+    "    dataset=\"BSC\",\n",
+    "    previous_scanpath=chinese_previous_scanpath,\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "399b583e",
+   "metadata": {},
+   "source": [
+    "### English Language / CELER Dataset"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "f6ae615b",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "keeping Bert with pre-trained weights\n"
+     ]
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "## English Raw-text generation"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "**Input:** The quick brown fox jumps over the lazy dog\n",
+       "\n",
+       "**Scanpath:** [0, 1, 4, 5, 8, 9, 3]\n",
+       "\n",
+       "**Fixated sequence:** The → fox → jumps → lazy → dog → brown\n",
+       "\n",
+       "| Fixation | Word / Character | Position | Source |\n",
+       "|---|---|---|---|\n",
+       "| 1 | The | 1 | Generated |\n",
+       "| 2 | fox | 4 | Generated |\n",
+       "| 3 | jumps | 5 | Generated |\n",
+       "| 4 | lazy | 8 | Generated |\n",
+       "| 5 | dog | 9 | Generated |\n",
+       "| 6 | brown | 3 | Generated |"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "---\n",
+       "\n",
+       "## English Continuation from a previous scanpath"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/markdown": [
+       "**Input:** The quick brown fox jumps over the lazy dog\n",
+       "\n",
+       "**Scanpath:** [0, 1, 2, 3, 6, 9, 10, 4, 5]\n",
+       "\n",
+       "**Fixated sequence:** The → quick → brown → over → dog\n",
+       "\n",
+       "| Fixation | Word / Character | Position | Source |\n",
+       "|---|---|---|---|\n",
+       "| 1 | The | 1 | Observed |\n",
+       "| 2 | quick | 2 | Observed |\n",
+       "| 3 | brown | 3 | Observed |\n",
+       "| 4 | over | 6 | Generated |\n",
+       "| 5 | dog | 9 | Generated |"
+      ],
+      "text/plain": [
+       "<IPython.core.display.Markdown object>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "english_text = \"The quick brown fox jumps over the lazy dog\"\n",
+    "english_previous_scanpath = [1,2,3]\n",
+    "\n",
+    "english_model = EyettentionRawTextInference(\n",
+    "    checkpoint_path=\"Eyettention/results/CELER/Eyettention_english.pth\",\n",
+    "    dataset=\"celer\",\n",
+    "    device=\"cpu\",\n",
+    ")\n",
+    "\n",
+    "def run_english_raw_text_input(model, text, max_pred_len=7):\n",
+    "    torch.manual_seed(42)\n",
+    "\n",
+    "    scanpath, _ = model.generate_from_english_text(\n",
+    "        text=text,\n",
+    "        max_pred_len=max_pred_len,\n",
+    "        previous_scanpath=None\n",
+    "    )\n",
+    "    \n",
+    "    return scanpath\n",
+    "\n",
+    "def run_english_previous_scanpath(model, text, max_pred_len=5, previous_scanpath=None):\n",
+    "    torch.manual_seed(42)\n",
+    "\n",
+    "    scanpath, _ = model.generate_from_english_text(\n",
+    "        text=text,\n",
+    "        max_pred_len=max_pred_len,\n",
+    "        previous_scanpath= previous_scanpath\n",
+    "    )\n",
+    "\n",
+    "    return scanpath\n",
+    "\n",
+    "\n",
+    "display(Markdown(\"## English Raw-text generation\"))\n",
+    "\n",
+    "raw_text = run_english_raw_text_input(english_model, english_text)\n",
+    "display_scanpath(english_text, scanpath=raw_text, dataset=\"celer\")\n",
+    "\n",
+    "\n",
+    "display(Markdown(\"---\\n\\n## English Continuation from a previous scanpath\"))\n",
+    "\n",
+    "\n",
+    "continuation = run_english_previous_scanpath(\n",
+    "    english_model,\n",
+    "    english_text,\n",
+    "    previous_scanpath=english_previous_scanpath,\n",
+    ")\n",
+    "\n",
+    "display_scanpath(english_text, scanpath=continuation, dataset=\"celer\", previous_scanpath=english_previous_scanpath)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "scanpaths (3.10.7)",
+   "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": "3.10.7"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}