chat template

.gitignore

@@ -1 +1,2 @@
 source
+local

cyber-chat.ipynb

@@ -0,0 +1,200 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "File Okay\n",
+      "Building strides for 8519 rows\n",
+      "Built 1221\n",
+      "Building strides for 2569 rows\n",
+      "Built 255\n",
+      "Building strides for 1260 rows\n",
+      "Built 165\n",
+      "Building strides for 4436 rows\n",
+      "Built 507\n",
+      "Building strides for 2968 rows\n",
+      "Built 260\n",
+      "2408\n"
+     ]
+    }
+   ],
+   "source": [
+    "import yaml\n",
+    "import json\n",
+    "import pandas as pd\n",
+    "import os\n",
+    "from typing import Literal\n",
+    "import random\n",
+    "\n",
+    "# We are going to move forward through each book, grabbing windows of text of context_length\n",
+    "# then, step forward by step_length, and grab the next window of text.\n",
+    "def build_strides(df : pd.DataFrame, context_length : int = 8192, step_length : int = 1024) -> list[str]: \n",
+    "    print(f\"Building strides for {len(df)} rows\")\n",
+    "    strides = []\n",
+    "    idx = 0\n",
+    "    stride = []\n",
+    "    words = 0\n",
+    "    ratio = context_length / step_length\n",
+    "    word_length = context_length * 3 / 4\n",
+    "    step_length = word_length / ratio\n",
+    "    chapter = None\n",
+    "    while True:\n",
+    "        if idx >= len(df):\n",
+    "            break\n",
+    "        if words > word_length:\n",
+    "            strides.append(stride.copy())\n",
+    "            # roll off entries from stride until we drop step_length words\n",
+    "            dropped = 0\n",
+    "            while dropped < step_length:\n",
+    "                dropped += stride[0][1]\n",
+    "                stride.pop(0)\n",
+    "            words -= dropped\n",
+    "        row = df.iloc[idx]\n",
+    "        my_chapter = row['chapter_ix']\n",
+    "        if my_chapter != chapter:\n",
+    "            stride.append(({'role': 'assistant', 'type': 'c', 'content': ''}, 0))\n",
+    "            chapter = my_chapter\n",
+    "        stride.append(({'role': 'assistant', 'type': 'p', 'content': row['text']}, row['word_count']))\n",
+    "        words += row['word_count']\n",
+    "        idx += 1\n",
+    "                    \n",
+    "    if len(stride) > 0:\n",
+    "        strides.append(stride)\n",
+    "\n",
+    "    print(f\"Built {len(strides)}\")\n",
+    "    return [[row[0] for row in stride] for stride in strides if len(stride) > 0]\n",
+    "\n",
+    "df = pd.read_parquet(\"./cyberpunk.parquet\")\n",
+    "print(\"File Okay\")\n",
+    "cyberbooks = ['cryptonomicon', 'neuromancer', 'burningchrome', 'snowcrash', 'monalisa']\n",
+    "samples = [{\"book\": book, \"message_ix\": i, \"messages\": line} for book in cyberbooks for i, line in enumerate(build_strides(df[df['book_name'] == book], 8192, 2048))]\n",
+    "print(len(samples))\n",
+    "\n",
+    "odf = pd.DataFrame(samples)\n",
+    "\n",
+    "datapath = \"./local\"\n",
+    "if os.path.exists(datapath) == False:\n",
+    "    os.makedirs(datapath)\n",
+    "odf.to_parquet('./local/cyber-chat.parquet', index=False)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/html": [
+       "<div>\n",
+       "<style scoped>\n",
+       "    .dataframe tbody tr th:only-of-type {\n",
+       "        vertical-align: middle;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe tbody tr th {\n",
+       "        vertical-align: top;\n",
+       "    }\n",
+       "\n",
+       "    .dataframe thead th {\n",
+       "        text-align: right;\n",
+       "    }\n",
+       "</style>\n",
+       "<table border=\"1\" class=\"dataframe\">\n",
+       "  <thead>\n",
+       "    <tr style=\"text-align: right;\">\n",
+       "      <th></th>\n",
+       "      <th>book</th>\n",
+       "      <th>message_ix</th>\n",
+       "      <th>messages</th>\n",
+       "    </tr>\n",
+       "  </thead>\n",
+       "  <tbody>\n",
+       "    <tr>\n",
+       "      <th>0</th>\n",
+       "      <td>cryptonomicon</td>\n",
+       "      <td>0</td>\n",
+       "      <td>[{'role': 'assistant', 'type': 'c', 'content':...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>1</th>\n",
+       "      <td>cryptonomicon</td>\n",
+       "      <td>1</td>\n",
+       "      <td>[{'role': 'assistant', 'type': 'p', 'content':...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>2</th>\n",
+       "      <td>cryptonomicon</td>\n",
+       "      <td>2</td>\n",
+       "      <td>[{'role': 'assistant', 'type': 'p', 'content':...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>3</th>\n",
+       "      <td>cryptonomicon</td>\n",
+       "      <td>3</td>\n",
+       "      <td>[{'role': 'assistant', 'type': 'p', 'content':...</td>\n",
+       "    </tr>\n",
+       "    <tr>\n",
+       "      <th>4</th>\n",
+       "      <td>cryptonomicon</td>\n",
+       "      <td>4</td>\n",
+       "      <td>[{'role': 'assistant', 'type': 'p', 'content':...</td>\n",
+       "    </tr>\n",
+       "  </tbody>\n",
+       "</table>\n",
+       "</div>"
+      ],
+      "text/plain": [
+       "            book  message_ix  \\\n",
+       "0  cryptonomicon           0   \n",
+       "1  cryptonomicon           1   \n",
+       "2  cryptonomicon           2   \n",
+       "3  cryptonomicon           3   \n",
+       "4  cryptonomicon           4   \n",
+       "\n",
+       "                                            messages  \n",
+       "0  [{'role': 'assistant', 'type': 'c', 'content':...  \n",
+       "1  [{'role': 'assistant', 'type': 'p', 'content':...  \n",
+       "2  [{'role': 'assistant', 'type': 'p', 'content':...  \n",
+       "3  [{'role': 'assistant', 'type': 'p', 'content':...  \n",
+       "4  [{'role': 'assistant', 'type': 'p', 'content':...  "
+      ]
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "odf.head()"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "txvenv",
+   "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": "3.10.12"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}