{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# How to count tokens with tiktoken\n",
    "\n",
    "[`tiktoken`](https://github.com/openai/tiktoken/blob/main/README.md) is a fast open-source tokenizer by OpenAI.\n",
    "\n",
    "Given a text string (e.g., `\"tiktoken is great!\"`) and an encoding (e.g., `\"cl100k_base\"`), a tokenizer can split the text string into a list of tokens (e.g., `[\"t\", \"ik\", \"token\", \" is\", \" great\", \"!\"]`).\n",
    "\n",
    "Splitting text strings into tokens is useful because GPT models see text in the form of tokens. Knowing how many tokens are in a text string can tell you (a) whether the string is too long for a text model to process and (b) how much an OpenAI API call costs (as usage is priced by token).\n",
    "\n",
    "\n",
    "## Encodings\n",
    "\n",
    "Encodings specify how text is converted into tokens. Different models use different encodings.\n",
    "\n",
    "`tiktoken` supports three encodings used by OpenAI models:\n",
    "\n",
    "| Encoding name           | OpenAI models                                       |\n",
    "|-------------------------|-----------------------------------------------------|\n",
    "| `cl100k_base`           | `gpt-4`, `gpt-3.5-turbo`, `text-embedding-ada-002`  |\n",
    "| `p50k_base`             | Codex models, `text-davinci-002`, `text-davinci-003`|\n",
    "| `r50k_base` (or `gpt2`) | GPT-3 models like `davinci`                         |\n",
    "\n",
    "You can retrieve the encoding for a model using `tiktoken.encoding_for_model()` as follows:\n",
    "```python\n",
    "encoding = tiktoken.encoding_for_model('gpt-3.5-turbo')\n",
    "```\n",
    "\n",
    "Note that `p50k_base` overlaps substantially with `r50k_base`, and for non-code applications, they will usually give the same tokens.\n",
    "\n",
    "## Tokenizer libraries by language\n",
    "\n",
    "For `cl100k_base` and `p50k_base` encodings:\n",
    "- Python: [tiktoken](https://github.com/openai/tiktoken/blob/main/README.md)\n",
    "- .NET / C#: [SharpToken](https://github.com/dmitry-brazhenko/SharpToken)\n",
    "- Java: [jtokkit](https://github.com/knuddelsgmbh/jtokkit)\n",
    "\n",
    "For `r50k_base` (`gpt2`) encodings, tokenizers are available in many languages.\n",
    "- Python: [tiktoken](https://github.com/openai/tiktoken/blob/main/README.md) (or alternatively [GPT2TokenizerFast](https://huggingface.co/docs/transformers/model_doc/gpt2#transformers.GPT2TokenizerFast))\n",
    "- JavaScript: [gpt-3-encoder](https://www.npmjs.com/package/gpt-3-encoder)\n",
    "- .NET / C#: [GPT Tokenizer](https://github.com/dluc/openai-tools)\n",
    "- Java: [gpt2-tokenizer-java](https://github.com/hyunwoongko/gpt2-tokenizer-java)\n",
    "- PHP: [GPT-3-Encoder-PHP](https://github.com/CodeRevolutionPlugins/GPT-3-Encoder-PHP)\n",
    "\n",
    "(OpenAI makes no endorsements or guarantees of third-party libraries.)\n",
    "\n",
    "\n",
    "## How strings are typically tokenized\n",
    "\n",
    "In English, tokens commonly range in length from one character to one word (e.g., `\"t\"` or `\" great\"`), though in some languages tokens can be shorter than one character or longer than one word. Spaces are usually grouped with the starts of words (e.g., `\" is\"` instead of `\"is \"` or `\" \"`+`\"is\"`). You can quickly check how a string is tokenized at the [OpenAI Tokenizer](https://beta.openai.com/tokenizer)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 0. Install `tiktoken`\n",
    "\n",
    "If needed, install `tiktoken` with `pip`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Requirement already satisfied: tiktoken in /Users/ted/.virtualenvs/openai/lib/python3.9/site-packages (0.3.2)\n",
      "Requirement already satisfied: regex>=2022.1.18 in /Users/ted/.virtualenvs/openai/lib/python3.9/site-packages (from tiktoken) (2022.10.31)\n",
      "Requirement already satisfied: requests>=2.26.0 in /Users/ted/.virtualenvs/openai/lib/python3.9/site-packages (from tiktoken) (2.28.2)\n",
      "Requirement already satisfied: charset-normalizer<4,>=2 in /Users/ted/.virtualenvs/openai/lib/python3.9/site-packages (from requests>=2.26.0->tiktoken) (2.0.9)\n",
      "Requirement already satisfied: idna<4,>=2.5 in /Users/ted/.virtualenvs/openai/lib/python3.9/site-packages (from requests>=2.26.0->tiktoken) (3.3)\n",
      "Requirement already satisfied: certifi>=2017.4.17 in /Users/ted/.virtualenvs/openai/lib/python3.9/site-packages (from requests>=2.26.0->tiktoken) (2021.10.8)\n",
      "Requirement already satisfied: urllib3<1.27,>=1.21.1 in /Users/ted/.virtualenvs/openai/lib/python3.9/site-packages (from requests>=2.26.0->tiktoken) (1.26.7)\n",
      "Note: you may need to restart the kernel to use updated packages.\n"
     ]
    }
   ],
   "source": [
    "%pip install --upgrade tiktoken"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Import `tiktoken`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tiktoken\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Load an encoding\n",
    "\n",
    "Use `tiktoken.get_encoding()` to load an encoding by name.\n",
    "\n",
    "The first time this runs, it will require an internet connection to download. Later runs won't need an internet connection."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "encoding = tiktoken.get_encoding(\"cl100k_base\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Use `tiktoken.encoding_for_model()` to automatically load the correct encoding for a given model name."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "encoding = tiktoken.encoding_for_model(\"gpt-3.5-turbo\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Turn text into tokens with `encoding.encode()`\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The `.encode()` method converts a text string into a list of token integers."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[83, 1609, 5963, 374, 2294, 0]"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "encoding.encode(\"tiktoken is great!\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Count tokens by counting the length of the list returned by `.encode()`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def num_tokens_from_string(string: str, encoding_name: str) -> int:\n",
    "    \"\"\"Returns the number of tokens in a text string.\"\"\"\n",
    "    encoding = tiktoken.get_encoding(encoding_name)\n",
    "    num_tokens = len(encoding.encode(string))\n",
    "    return num_tokens\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "6"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_tokens_from_string(\"tiktoken is great!\", \"cl100k_base\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Turn tokens into text with `encoding.decode()`"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "`.decode()` converts a list of token integers to a string."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'tiktoken is great!'"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "encoding.decode([83, 1609, 5963, 374, 2294, 0])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Warning: although `.decode()` can be applied to single tokens, beware that it can be lossy for tokens that aren't on utf-8 boundaries."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For single tokens, `.decode_single_token_bytes()` safely converts a single integer token to the bytes it represents."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[b't', b'ik', b'token', b' is', b' great', b'!']"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "[encoding.decode_single_token_bytes(token) for token in [83, 1609, 5963, 374, 2294, 0]]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "(The `b` in front of the strings indicates that the strings are byte strings.)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 5. Comparing encodings\n",
    "\n",
    "Different encodings vary in how they split words, group spaces, and handle non-English characters. Using the methods above, we can compare different encodings on a few example strings."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def compare_encodings(example_string: str) -> None:\n",
    "    \"\"\"Prints a comparison of three string encodings.\"\"\"\n",
    "    # print the example string\n",
    "    print(f'\\nExample string: \"{example_string}\"')\n",
    "    # for each encoding, print the # of tokens, the token integers, and the token bytes\n",
    "    for encoding_name in [\"gpt2\", \"p50k_base\", \"cl100k_base\"]:\n",
    "        encoding = tiktoken.get_encoding(encoding_name)\n",
    "        token_integers = encoding.encode(example_string)\n",
    "        num_tokens = len(token_integers)\n",
    "        token_bytes = [encoding.decode_single_token_bytes(token) for token in token_integers]\n",
    "        print()\n",
    "        print(f\"{encoding_name}: {num_tokens} tokens\")\n",
    "        print(f\"token integers: {token_integers}\")\n",
    "        print(f\"token bytes: {token_bytes}\")\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Example string: \"antidisestablishmentarianism\"\n",
      "\n",
      "gpt2: 5 tokens\n",
      "token integers: [415, 29207, 44390, 3699, 1042]\n",
      "token bytes: [b'ant', b'idis', b'establishment', b'arian', b'ism']\n",
      "\n",
      "p50k_base: 5 tokens\n",
      "token integers: [415, 29207, 44390, 3699, 1042]\n",
      "token bytes: [b'ant', b'idis', b'establishment', b'arian', b'ism']\n",
      "\n",
      "cl100k_base: 6 tokens\n",
      "token integers: [519, 85342, 34500, 479, 8997, 2191]\n",
      "token bytes: [b'ant', b'idis', b'establish', b'ment', b'arian', b'ism']\n"
     ]
    }
   ],
   "source": [
    "compare_encodings(\"antidisestablishmentarianism\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Example string: \"2 + 2 = 4\"\n",
      "\n",
      "gpt2: 5 tokens\n",
      "token integers: [17, 1343, 362, 796, 604]\n",
      "token bytes: [b'2', b' +', b' 2', b' =', b' 4']\n",
      "\n",
      "p50k_base: 5 tokens\n",
      "token integers: [17, 1343, 362, 796, 604]\n",
      "token bytes: [b'2', b' +', b' 2', b' =', b' 4']\n",
      "\n",
      "cl100k_base: 7 tokens\n",
      "token integers: [17, 489, 220, 17, 284, 220, 19]\n",
      "token bytes: [b'2', b' +', b' ', b'2', b' =', b' ', b'4']\n"
     ]
    }
   ],
   "source": [
    "compare_encodings(\"2 + 2 = 4\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Example string: \"お誕生日おめでとう\"\n",
      "\n",
      "gpt2: 14 tokens\n",
      "token integers: [2515, 232, 45739, 243, 37955, 33768, 98, 2515, 232, 1792, 223, 30640, 30201, 29557]\n",
      "token bytes: [b'\\xe3\\x81', b'\\x8a', b'\\xe8\\xaa', b'\\x95', b'\\xe7\\x94\\x9f', b'\\xe6\\x97', b'\\xa5', b'\\xe3\\x81', b'\\x8a', b'\\xe3\\x82', b'\\x81', b'\\xe3\\x81\\xa7', b'\\xe3\\x81\\xa8', b'\\xe3\\x81\\x86']\n",
      "\n",
      "p50k_base: 14 tokens\n",
      "token integers: [2515, 232, 45739, 243, 37955, 33768, 98, 2515, 232, 1792, 223, 30640, 30201, 29557]\n",
      "token bytes: [b'\\xe3\\x81', b'\\x8a', b'\\xe8\\xaa', b'\\x95', b'\\xe7\\x94\\x9f', b'\\xe6\\x97', b'\\xa5', b'\\xe3\\x81', b'\\x8a', b'\\xe3\\x82', b'\\x81', b'\\xe3\\x81\\xa7', b'\\xe3\\x81\\xa8', b'\\xe3\\x81\\x86']\n",
      "\n",
      "cl100k_base: 9 tokens\n",
      "token integers: [33334, 45918, 243, 21990, 9080, 33334, 62004, 16556, 78699]\n",
      "token bytes: [b'\\xe3\\x81\\x8a', b'\\xe8\\xaa', b'\\x95', b'\\xe7\\x94\\x9f', b'\\xe6\\x97\\xa5', b'\\xe3\\x81\\x8a', b'\\xe3\\x82\\x81', b'\\xe3\\x81\\xa7', b'\\xe3\\x81\\xa8\\xe3\\x81\\x86']\n"
     ]
    }
   ],
   "source": [
    "compare_encodings(\"お誕生日おめでとう\")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 6. Counting tokens for chat API calls\n",
    "\n",
    "ChatGPT models like `gpt-3.5-turbo` and `gpt-4` use tokens in the same way as older completions models, but because of their message-based formatting, it's more difficult to count how many tokens will be used by a conversation.\n",
    "\n",
    "Below is an example function for counting tokens for messages passed to `gpt-3.5-turbo-0301` or `gpt-4-0314`.\n",
    "\n",
    "Note that the exact way that tokens are counted from messages may change from model to model. Consider the counts from the function below an estimate, not a timeless guarantee."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def num_tokens_from_messages(messages, model=\"gpt-3.5-turbo-0301\"):\n",
    "    \"\"\"Returns the number of tokens used by a list of messages.\"\"\"\n",
    "    try:\n",
    "        encoding = tiktoken.encoding_for_model(model)\n",
    "    except KeyError:\n",
    "        print(\"Warning: model not found. Using cl100k_base encoding.\")\n",
    "        encoding = tiktoken.get_encoding(\"cl100k_base\")\n",
    "    if model == \"gpt-3.5-turbo\":\n",
    "        print(\"Warning: gpt-3.5-turbo may change over time. Returning num tokens assuming gpt-3.5-turbo-0301.\")\n",
    "        return num_tokens_from_messages(messages, model=\"gpt-3.5-turbo-0301\")\n",
    "    elif model == \"gpt-4\":\n",
    "        print(\"Warning: gpt-4 may change over time. Returning num tokens assuming gpt-4-0314.\")\n",
    "        return num_tokens_from_messages(messages, model=\"gpt-4-0314\")\n",
    "    elif model == \"gpt-3.5-turbo-0301\":\n",
    "        tokens_per_message = 4  # every message follows <|start|>{role/name}\\n{content}<|end|>\\n\n",
    "        tokens_per_name = -1  # if there's a name, the role is omitted\n",
    "    elif model == \"gpt-4-0314\":\n",
    "        tokens_per_message = 3\n",
    "        tokens_per_name = 1\n",
    "    else:\n",
    "        raise NotImplementedError(f\"\"\"num_tokens_from_messages() is not implemented for model {model}. See https://github.com/openai/openai-python/blob/main/chatml.md for information on how messages are converted to tokens.\"\"\")\n",
    "    num_tokens = 0\n",
    "    for message in messages:\n",
    "        num_tokens += tokens_per_message\n",
    "        for key, value in message.items():\n",
    "            num_tokens += len(encoding.encode(value))\n",
    "            if key == \"name\":\n",
    "                num_tokens += tokens_per_name\n",
    "    num_tokens += 3  # every reply is primed with <|start|>assistant<|message|>\n",
    "    return num_tokens\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "gpt-3.5-turbo-0301\n",
      "127 prompt tokens counted by num_tokens_from_messages().\n",
      "127 prompt tokens counted by the OpenAI API.\n",
      "\n",
      "gpt-4-0314\n",
      "129 prompt tokens counted by num_tokens_from_messages().\n",
      "129 prompt tokens counted by the OpenAI API.\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# let's verify the function above matches the OpenAI API response\n",
    "\n",
    "import openai\n",
    "\n",
    "example_messages = [\n",
    "    {\n",
    "        \"role\": \"system\",\n",
    "        \"content\": \"You are a helpful, pattern-following assistant that translates corporate jargon into plain English.\",\n",
    "    },\n",
    "    {\n",
    "        \"role\": \"system\",\n",
    "        \"name\": \"example_user\",\n",
    "        \"content\": \"New synergies will help drive top-line growth.\",\n",
    "    },\n",
    "    {\n",
    "        \"role\": \"system\",\n",
    "        \"name\": \"example_assistant\",\n",
    "        \"content\": \"Things working well together will increase revenue.\",\n",
    "    },\n",
    "    {\n",
    "        \"role\": \"system\",\n",
    "        \"name\": \"example_user\",\n",
    "        \"content\": \"Let's circle back when we have more bandwidth to touch base on opportunities for increased leverage.\",\n",
    "    },\n",
    "    {\n",
    "        \"role\": \"system\",\n",
    "        \"name\": \"example_assistant\",\n",
    "        \"content\": \"Let's talk later when we're less busy about how to do better.\",\n",
    "    },\n",
    "    {\n",
    "        \"role\": \"user\",\n",
    "        \"content\": \"This late pivot means we don't have time to boil the ocean for the client deliverable.\",\n",
    "    },\n",
    "]\n",
    "\n",
    "for model in [\"gpt-3.5-turbo-0301\", \"gpt-4-0314\"]:\n",
    "    print(model)\n",
    "    # example token count from the function defined above\n",
    "    print(f\"{num_tokens_from_messages(example_messages, model)} prompt tokens counted by num_tokens_from_messages().\")\n",
    "    # example token count from the OpenAI API\n",
    "    response = openai.ChatCompletion.create(\n",
    "        model=model,\n",
    "        messages=example_messages,\n",
    "        temperature=0,\n",
    "        max_tokens=1  # we're only counting input tokens here, so let's not waste tokens on the output\n",
    "    )\n",
    "    print(f'{response[\"usage\"][\"prompt_tokens\"]} prompt tokens counted by the OpenAI API.')\n",
    "    print()\n"
   ]
  }
 ],
 "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": "3.7.3"
  },
  "vscode": {
   "interpreter": {
    "hash": "365536dcbde60510dc9073d6b991cd35db2d9bac356a11f5b64279a5e6708b97"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}