2023_07_08_mpt-30b-chat_langchain_vastai.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "MIT License\n",
    "\n",
    "Copyright (c) 2023 Pavel Shibanov [https://blog.experienced.dev/](https://blog.experienced.dev/?utm_source=notebooks)\n",
    "\n",
    "Permission is hereby granted, free of charge, to any person obtaining a copy\n",
    "of this software and associated documentation files (the \"Software\"), to deal\n",
    "in the Software without restriction, including without limitation the rights\n",
    "to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n",
    "copies of the Software, and to permit persons to whom the Software is\n",
    "furnished to do so, subject to the following conditions:\n",
    "\n",
    "The above copyright notice and this permission notice shall be included in all\n",
    "copies or substantial portions of the Software.\n",
    "\n",
    "THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n",
    "IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n",
    "FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n",
    "AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n",
    "LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n",
    "OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\n",
    "SOFTWARE."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# [LangChain](https://github.com/hwchase17/langchain) + [MPT-30B-Chat](https://huggingface.co/mosaicml/mpt-30b-chat) with [Text Generation Inference](https://github.com/huggingface/text-generation-inference) on [vast.ai](https://cloud.vast.ai/?ref=71973) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/jpeg": "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",
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"560\"\n",
       "            height=\"315\"\n",
       "            src=\"https://www.youtube.com/embed/592CoLjNZQ0\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "            \n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.YouTubeVideo at 0x7f4a56793d60>"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from IPython.display import YouTubeVideo\n",
    "\n",
    "YouTubeVideo(\"592CoLjNZQ0\", width=\"560\", height=\"315\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Installing the dependencies\n",
    "%pip install -q python-dotenv==1.0.0 langchain==0.0.226 text-generation==0.6.0 httpx==0.24.1 gradio==3.36.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Load .env file\n",
    "from dotenv import load_dotenv, find_dotenv\n",
    "\n",
    "_ = load_dotenv(find_dotenv())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Download the [vast.ai](https://cloud.vast.ai/?ref=71973) CLI\n",
    "import os\n",
    "import urllib.request\n",
    "\n",
    "urllib.request.urlretrieve(\n",
    "    \"https://raw.githubusercontent.com/vast-ai/vast-python/7d2f10f/vast.py\", \"vast\"\n",
    ")\n",
    "os.chmod(\"vast\", 0o755)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Helper to extract dictionaries from\n",
    "# @markdown [vast.ai CLI](https://cloud.vast.ai/?ref=71973)\n",
    "import subprocess\n",
    "import json\n",
    "\n",
    "\n",
    "def vast_ai_cmd(args=None):\n",
    "    if not args:\n",
    "        args = [\"--help\"]\n",
    "    cmd = [\"./vast\"]\n",
    "    cmd += args\n",
    "    result = subprocess.run(cmd, capture_output=True)\n",
    "    if result.stderr:\n",
    "        print(result.stderr)\n",
    "    output = result.stdout.decode()\n",
    "\n",
    "    if \"--raw\" in cmd:\n",
    "        try:\n",
    "            data = json.loads(output)\n",
    "            return data\n",
    "        except json.JSONDecodeError as e:\n",
    "            print(f\"An error occurred: {str(e)}\")\n",
    "    else:\n",
    "        print(output)\n",
    "\n",
    "\n",
    "vast_ai_cmd()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown vast.ai API key. You can obtain one at [vast.ai](https://cloud.vast.ai/?ref=71973)\n",
    "import os\n",
    "from IPython.display import display, Markdown\n",
    "\n",
    "VAST_AI_API_KEY = os.getenv(\"VAST_AI_API_KEY\", \"your_vast_ai_api_key\")\n",
    "if VAST_AI_API_KEY == \"your_vast_ai_api_key\":\n",
    "    display(\n",
    "        Markdown(\n",
    "            \"It appears that you don't have a vast.ai API key. You can obtain one at [vast.ai](https://cloud.vast.ai/?ref=71973)\"\n",
    "        )\n",
    "    )\n",
    "    raise AssertionError(\"Missing vast.ai API key\")\n",
    "\n",
    "vast_ai_cmd([\"set\", \"api-key\", VAST_AI_API_KEY])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Searching for suitable hardware to run mosaicml/mpt-30b-chat with quantization.\n",
    "import pandas as pd\n",
    "\n",
    "raw_offers = vast_ai_cmd(\n",
    "    [\n",
    "        \"search\",\n",
    "        \"offers\",\n",
    "        \"--raw\",\n",
    "        \"reliability > 0.9 cuda_vers >= 11.8 num_gpus = 1 gpu_ram >= 47 inet_down >= 500 disk_space >= 130\",\n",
    "        \"-o\",\n",
    "        \"dph\",\n",
    "    ]\n",
    ")\n",
    "\n",
    "offers_count = len(raw_offers)\n",
    "if offers_count:\n",
    "    print(f\"{offers_count=}\")\n",
    "    offers = pd.DataFrame.from_dict(raw_offers)\n",
    "else:\n",
    "    raise AssertionError(\"No offers, change search query\")\n",
    "offers[\n",
    "    [\n",
    "        \"id\",\n",
    "        \"dph_total\",\n",
    "        \"inet_down\",\n",
    "        \"cuda_max_good\",\n",
    "        \"num_gpus\",\n",
    "        \"gpu_name\",\n",
    "        \"gpu_ram\",\n",
    "        \"cpu_ram\",\n",
    "        \"disk_space\",\n",
    "    ]\n",
    "].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Selecting a hardware.\n",
    "selected_id = 0\n",
    "selected = offers.iloc[selected_id]\n",
    "selected.id, selected.dph_total"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Starting [Text Generation Inference](https://github.com/huggingface/text-generation-inference) with [MPT-30B-Chat](https://huggingface.co/mosaicml/mpt-30b-chat) model.\n",
    "res = vast_ai_cmd(\n",
    "    [\n",
    "        \"create\",\n",
    "        \"instance\",\n",
    "        \"--raw\",\n",
    "        str(selected.id),\n",
    "        \"--env\",\n",
    "        \"-p 80:80\",\n",
    "        \"--disk\",\n",
    "        \"130\",\n",
    "        \"--image\",\n",
    "        \"ghcr.io/huggingface/text-generation-inference:0.9.1\",\n",
    "        \"--args\",\n",
    "        \"--model-id\",\n",
    "        \"mosaicml/mpt-30b-chat\",\n",
    "        \"--quantize\",\n",
    "        \"bitsandbytes\",\n",
    "        \"--hostname\",\n",
    "        \"::\",\n",
    "    ]\n",
    ")\n",
    "instance_id = res[\"new_contract\"]\n",
    "instance_id"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Waiting for instance to become running.\n",
    "import time\n",
    "from IPython.display import clear_output\n",
    "\n",
    "is_running = False\n",
    "while not is_running:\n",
    "    instances = vast_ai_cmd([\"show\", \"instances\", \"--raw\"])\n",
    "    instance = next((i for i in instances if i[\"id\"] == instance_id), None)\n",
    "    is_running = instance[\"actual_status\"] == \"running\"\n",
    "    clear_output(wait=True)\n",
    "    print(instance)\n",
    "    time.sleep(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Fetching the inference_server_url from instance configuration.\n",
    "hostname = instance[\"public_ipaddr\"].strip()\n",
    "port = instance[\"ports\"][\"80/tcp\"][-1][\"HostPort\"]\n",
    "inference_server_url = f\"http://{hostname}:{port}\"\n",
    "inference_server_url"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Waiting for the TGI health check and displaying logs every 5 seconds.\n",
    "from IPython.display import clear_output\n",
    "import time\n",
    "import httpx\n",
    "\n",
    "url = f\"{inference_server_url}/health\"\n",
    "is_healthy = False\n",
    "while not is_healthy:\n",
    "    try:\n",
    "        res = httpx.get(url)\n",
    "        is_healthy = res.status_code == 200\n",
    "    except:\n",
    "        clear_output(wait=True)\n",
    "        vast_ai_cmd([\"logs\", str(instance_id)])\n",
    "    time.sleep(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Creating llm with LangChain's HuggingFaceTextGenInference\n",
    "from langchain.llms import HuggingFaceTextGenInference\n",
    "\n",
    "llm = HuggingFaceTextGenInference(\n",
    "    inference_server_url=inference_server_url,\n",
    "    max_new_tokens=1024,\n",
    "    top_k=10,\n",
    "    top_p=0.95,\n",
    "    typical_p=0.95,\n",
    "    stop_sequences=[\n",
    "        \"Human:\",\n",
    "        \"AI:\",\n",
    "    ],  # If 'stop_sequences' is not present, the model will generate both sides of the conversation.\n",
    "    temperature=0.1,\n",
    "    repetition_penalty=1.03,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Creating chat memory.\n",
    "from langchain.chains.conversation.memory import ConversationBufferWindowMemory\n",
    "\n",
    "memory = ConversationBufferWindowMemory(\n",
    "    memory_key=\"history\",\n",
    "    k=6,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Creating ConversationChain.\n",
    "from langchain.chains import ConversationChain\n",
    "\n",
    "chat = ConversationChain(llm=llm, memory=memory, verbose=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Chat prompt.\n",
    "chat.prompt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Chat prompt template.\n",
    "display(Markdown(chat.prompt.template))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown 1st message.\n",
    "display(Markdown(chat.predict(input=\"hi how are you?\")))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown 2nd message.\n",
    "display(Markdown(chat.predict(input=\"My name is Pavel.\")))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Check memory state.\n",
    "chat.memory"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Check if memory is functioning.\n",
    "display(Markdown(chat.predict(input=\"What is my name?\")))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Next question.\n",
    "display(\n",
    "    Markdown(\n",
    "        chat.predict(input=\"What is the difference between a method and a function?\")\n",
    "    )\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Follow up question.\n",
    "display(Markdown(chat.predict(input=\"What is more common in OOP?\")))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Starting Gradio app with the cachatbot.\n",
    "import random\n",
    "import time\n",
    "import gradio as gr\n",
    "\n",
    "from langchain.chains.conversation.memory import ConversationBufferWindowMemory\n",
    "\n",
    "\n",
    "memory = ConversationBufferWindowMemory(\n",
    "    memory_key=\"history\",\n",
    "    k=6,\n",
    ")\n",
    "chat = ConversationChain(llm=llm, memory=memory)\n",
    "\n",
    "chatbot = gr.Chatbot()\n",
    "human_message = gr.Textbox()\n",
    "\n",
    "\n",
    "def respond(human_message, chat_history):\n",
    "    ai_message = chat.predict(input=human_message)\n",
    "    chat_history.append((human_message, ai_message))\n",
    "    return \"\", chat_history\n",
    "\n",
    "\n",
    "demo = gr.Interface(\n",
    "    fn=respond,\n",
    "    examples=[\n",
    "        [\"How to square a circle?\"],\n",
    "        [\"Compare the physical and chemical properties of hydrogen and oxygen\"],\n",
    "        [\"What is the length of human DNA in meters?\"],\n",
    "    ],\n",
    "    inputs=[human_message, chatbot],\n",
    "    outputs=[human_message, chatbot],\n",
    "    allow_flagging=\"never\",\n",
    ")\n",
    "try:\n",
    "    import google.colab\n",
    "\n",
    "    IN_COLAB = True\n",
    "except:\n",
    "    IN_COLAB = False\n",
    "demo.launch(demo.launch(share=IN_COLAB, debug=True))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Stop the instance.\n",
    "vast_ai_cmd([\"stop\", \"instance\", str(instance_id)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Start the instance.\n",
    "vast_ai_cmd([\"start\", \"instance\", str(instance_id)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Destroy the instance.\n",
    "vast_ai_cmd([\"destroy\", \"instance\", str(instance_id)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# @markdown Confirm that you are not charged.\n",
    "vast_ai_cmd([\"show\", \"instances\", \"--raw\"])"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "langchain-tutorial",
   "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": "3.10.10"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}