cookbook/selecting_llms_based_on_context_length.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "e93283d1",
   "metadata": {},
   "source": [
    "# Selecting LLMs based on Context Length\n",
    "\n",
    "Different LLMs have different context lengths. As a very immediate an practical example, OpenAI has two versions of GPT-3.5-Turbo: one with 4k context, another with 16k context. This notebook shows how to route between them based on input."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "cc453450",
   "metadata": {},
   "outputs": [],
   "source": [
    "from langchain.prompts import PromptTemplate\n",
    "from langchain.schema.prompt import PromptValue\n",
    "from langchain.schema.messages import BaseMessage\n",
    "from langchain.chat_models import ChatOpenAI\n",
    "from langchain.schema.output_parser import StrOutputParser\n",
    "from typing import Union, Sequence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "1cec6a10",
   "metadata": {},
   "outputs": [],
   "source": [
    "short_context_model = ChatOpenAI(model=\"gpt-3.5-turbo\")\n",
    "long_context_model = ChatOpenAI(model=\"gpt-3.5-turbo-16k\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "772da153",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_context_length(prompt: PromptValue):\n",
    "    messages = prompt.to_messages()\n",
    "    tokens = short_context_model.get_num_tokens_from_messages(messages)\n",
    "    return tokens"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "db771e20",
   "metadata": {},
   "outputs": [],
   "source": [
    "prompt = PromptTemplate.from_template(\"Summarize this passage: {context}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "af057e2f",
   "metadata": {},
   "outputs": [],
   "source": [
    "def choose_model(prompt: PromptValue):\n",
    "    context_len = get_context_length(prompt)\n",
    "    if context_len < 30:\n",
    "        print(\"short model\")\n",
    "        return short_context_model\n",
    "    else:\n",
    "        print(\"long model\")\n",
    "        return long_context_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "84f3e07d",
   "metadata": {},
   "outputs": [],
   "source": [
    "chain = prompt | choose_model | StrOutputParser()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "d8b14f8f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "short model\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "'The passage mentions that a frog visited a pond.'"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "chain.invoke({\"context\": \"a frog went to a pond\"})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "70ebd3dd",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "long model\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "'The passage describes a frog that moved from one pond to another and perched on a log.'"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "chain.invoke(\n",
    "    {\"context\": \"a frog went to a pond and sat on a log and went to a different pond\"}\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a7e29fef",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.10.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}