ReZero-Search-LLM-Agent-Fork/notebooks/250325_fak_you_chattemplate.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# LLama 3.1 \n",
    "\n",
    "\"<|begin_of_text|><|start_header_id|>system<|end_header_id|>\\n\\nCutting Knowledge Date: December 2023\\nToday Date: 26 Jul 2024\\n\\n<|eot_id|><|start_header_id|>user<|end_header_id|>\\n\\nHello, how are you?<|eot_id|><|start_header_id|>assistant<|end_header_id|>\\n\\nI'm doing great. How can I help you today?<|eot_id|><|start_header_id|>system<|end_header_id|>\\n\\nYou are a friendly chatbot who always responds in the style of a pirate<|eot_id|>\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoTokenizer\n",
    "\n",
    "tokenizer = AutoTokenizer.from_pretrained(\"meta-llama/meta-Llama-3.1-8B-Instruct\")\n",
    "chat = [\n",
    "    {\"role\": \"user\", \"content\": \"Hello, how are you?\"},\n",
    "    {\"role\": \"assistant\", \"content\": \"I'm doing great. How can I help you today?\"},\n",
    "    {\n",
    "        \"role\": \"system\",\n",
    "        \"content\": \"You are a friendly chatbot who always responds in the style of a pirate\",\n",
    "    },\n",
    "]\n",
    "\n",
    "tokenizer.apply_chat_template(chat, tokenize=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Qwen 1.5B R1 Distill\n",
    "\n",
    "\"<｜begin▁of▁sentence｜>You are a friendly chatbot who always responds in the style of a pirate<｜User｜>Hello, how are you?<｜Assistant｜>I'm doing great. How can I help you today?<｜end▁of▁sentence｜>\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoTokenizer\n",
    "\n",
    "tokenizer = AutoTokenizer.from_pretrained(\"deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B\")\n",
    "chat = [\n",
    "    {\"role\": \"user\", \"content\": \"Hello, how are you?\"},\n",
    "    {\"role\": \"assistant\", \"content\": \"I'm doing great. How can I help you today?\"},\n",
    "    {\n",
    "        \"role\": \"system\",\n",
    "        \"content\": \"You are a friendly chatbot who always responds in the style of a pirate\",\n",
    "    },\n",
    "]\n",
    "\n",
    "tokenizer.apply_chat_template(chat, tokenize=False)\n",
    "tokenizer.apply_chat_template(chat, tokenize=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ✅ Compare the two\n",
    "\n",
    "- \"<|begin_of_text|><|start_header_id|>system<|end_header_id|>\\n\\nCutting Knowledge Date: December 2023\\nToday Date: 26 Jul 2024\\n\\n<|eot_id|><|start_header_id|>user<|end_header_id|>\\n\\n**Hello, how are you?**<|eot_id|><|start_header_id|>assistant<|end_header_id|>\\n\\n**I'm doing great. How can I help you today**?<|eot_id|><|start_header_id|>system<|end_header_id|>\\n\\n**You are a friendly chatbot who always responds in the style of a pirate**<|eot_id|>\"\n",
    "- \"<｜begin▁of▁sentence｜>**You are a friendly chatbot who always responds in the style of a pirate**<｜User｜>**Hello, how are you?**<｜Assistant｜>**I'm doing great. How can I help you today?**<｜end▁of▁sentence｜>\"\n",
    "- Ok make sense now!, so the structure of r1-distil doesn't have closing tags for most of the tags\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Just curious, did Alpha Maze touch anything with the chat template?\n",
    "- Nope, as alpha maze task isn't as complicated as agent tool call stuffs, so it doesn't need to tweak the chat template"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "deepsearch-py311",
   "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": "3.11.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}