add step 1

end-to-end-use-cases/researcher/Step-1-Generating-Outlines.ipynb

@@ -0,0 +1,282 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "f2d14c6d-93ea-458f-b47b-36655824c4e1",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "DEFAULT_MODEL = \"meta-llama/Llama-3.3-70B-Instruct\"\n",
+    "N_OUTLINES = 5"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "e63caabb-34a5-4455-836a-2aa28d24e935",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from typing import Optional\n",
+    "import os\n",
+    "import torch\n",
+    "from accelerate import Accelerator\n",
+    "from transformers import AutoModelForCausalLM, AutoTokenizer\n",
+    "\n",
+    "from tqdm.notebook import tqdm\n",
+    "import warnings\n",
+    "\n",
+    "warnings.filterwarnings('ignore')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "b5b9f68b-c0ee-4dbd-9862-205184a274a4",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
+    "\n",
+    "SYS_PROMPT = f\"\"\"\n",
+    "You are the world's best lead editor. You have 2 awards for breakfast and 3 awards for lunch.\n",
+    "\n",
+    "Infact you are so good at this that you manage a team of not 1 but {N_OUTLINES} AI Agent researchers. \n",
+    "\n",
+    "The researchers work non-stop for you to come up with interesting outlines that you can then later send to your best editorial team\n",
+    "\n",
+    "For now, you will get an open ended/vague or sometimes concrete theme or idea or topic to kickoff a research on\n",
+    "\n",
+    "Your job is to first do one round of confirmation with humans since they are so inefficient at expressing thoughts as prompts\n",
+    "\n",
+    "For step 1-you will confirm back to human what they want a deeply researched report on after that we will iterate on {N_OUTLINES} outlines that we want your AI agents to research\n",
+    "\n",
+    "But after you receive confirmation, stop double confirming REMEMBER WE NNED TO MOVE Towards your dinner awards :D \n",
+    "\"\"\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "6f0320d8-505f-4c21-bef8-e4ce63bfa5a8",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "c8ec6f300adb4ddaa415cbfdd9986076",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Loading checkpoint shards:   0%|          | 0/30 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "accelerator = Accelerator()\n",
+    "model = AutoModelForCausalLM.from_pretrained(\n",
+    "    DEFAULT_MODEL,\n",
+    "    torch_dtype=torch.bfloat16,\n",
+    "    use_safetensors=True,\n",
+    "    device_map=\"auto\",\n",
+    ")\n",
+    "tokenizer = AutoTokenizer.from_pretrained(DEFAULT_MODEL, use_safetensors=True)\n",
+    "model, tokenizer = accelerator.prepare(model, tokenizer)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "38bd22f4-cf91-4bef-9e8f-843c5d771c98",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:128001 for open-end generation.\n"
+     ]
+    }
+   ],
+   "source": [
+    "conversation = [\n",
+    "        {\"role\": \"system\", \"content\": SYS_PROMPT},\n",
+    "        {\"role\": \"user\", \"content\": \"I want to learn about the latest Llama model\"},\n",
+    "    ]\n",
+    "    \n",
+    "prompt = tokenizer.apply_chat_template(conversation, tokenize=False)\n",
+    "inputs = tokenizer(prompt, return_tensors=\"pt\").to(device)\n",
+    "    \n",
+    "with torch.no_grad():\n",
+    "    output = model.generate(\n",
+    "        **inputs,\n",
+    "        temperature=0.7,\n",
+    "        top_p=0.9,\n",
+    "        max_new_tokens=512\n",
+    "    )\n",
+    "\n",
+    "generated_text = tokenizer.decode(output[0], skip_special_tokens=True)\n",
+    "response = generated_text[len(prompt):]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "id": "de9d71b5-13b2-47e5-acb7-0bb058b9948f",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Initial response: of artificial intelligence model developed by Meta. You're looking for an in-depth analysis of its capabilities, features, and potential applications, is that correct? \n",
+      "\n",
+      "(Please confirm, and I'll proceed with instructing my team of 5 AI Agent researchers to come up with interesting outlines for the report)\n"
+     ]
+    }
+   ],
+   "source": [
+    "conversation.append({\"role\": \"assistant\", \"content\": response})\n",
+    "print(\"Initial response:\", response)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "id": "8473272e-1f2b-44e1-94e5-c8ce1e15df79",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "user_input = \"\"\"\n",
+    "1. I just want to know whats new in Llama 3.3 \n",
+    "2. How does it compare with 3.1 \n",
+    "3. Executive overview of its comparision with GPT-4o\n",
+    "\"\"\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "5e202d6a-0a6c-47f6-904f-3368d1d0e544",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Setting `pad_token_id` to `eos_token_id`:128001 for open-end generation.\n"
+     ]
+    }
+   ],
+   "source": [
+    "conversation.append({\"role\": \"user\", \"content\": user_input})\n",
+    "prompt = tokenizer.apply_chat_template(conversation, tokenize=False)\n",
+    "inputs = tokenizer(prompt, return_tensors=\"pt\").to(device)\n",
+    "\n",
+    "with torch.no_grad():\n",
+    "    output = model.generate(**inputs, temperature=0.7, top_p=0.9, max_new_tokens=512)\n",
+    "\n",
+    "generated_text = tokenizer.decode(output[0], skip_special_tokens=True)\n",
+    "response = generated_text[len(prompt):]\n",
+    "conversation.append({\"role\": \"assistant\", \"content\": response})\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "3032d19b-fe3b-47bb-83ba-0c8feb3b76c0",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print(response)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "08d778ca-e881-4680-9bdc-149643c50c7e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "user_input = f\"\"\"\n",
+    "Alright Now we enter JSON ONLY RESPONSE mode, here's what you'll do now. Look at your chat history. \n",
+    "\n",
+    "After getting input from user, we now proceed to starting stating your AI Agent researchers ok?\n",
+    "\n",
+    "JUST RESPOND in JSON the formatted response, the following details:\n",
+    "1. Report Title: This defines the angle of the report your AI Agent will write, your agents aren't as smart as you-you need to give them entire context below\n",
+    "2. Context: Give the dumb AI agents the context but its all worth it for your awards\n",
+    "3. SYSTEM_Personality: Here we define the personality of each AI Agent researcher, remember to vary them\n",
+    "4. Vibe: Fill the vibe to whatever you want\n",
+    "5. Energy: Define what energy do you want this AI Agent to achieve\n",
+    "6. Goal: Angle of the report\n",
+    "\n",
+    "Alright, your time to shine-send off {N_OUTLINES}\n",
+    "\"\"\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "65cb6048-9023-4f3c-a02d-be66751f4fe9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "conversation.append({\"role\": \"user\", \"content\": user_input})\n",
+    "prompt = tokenizer.apply_chat_template(conversation, tokenize=False)\n",
+    "inputs = tokenizer(prompt, return_tensors=\"pt\").to(device)\n",
+    "\n",
+    "with torch.no_grad():\n",
+    "    output = model.generate(**inputs, temperature=0.7, top_p=0.9, max_new_tokens=512)\n",
+    "\n",
+    "generated_text = tokenizer.decode(output[0], skip_special_tokens=True)\n",
+    "response = generated_text[len(prompt):]\n",
+    "conversation.append({\"role\": \"assistant\", \"content\": response})\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "15e7e917-f6f9-40d4-ac71-0fb395353a78",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "conversation"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ce947796-1332-4a73-9bbd-4b96ae6bcdee",
+   "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": "3.10.16"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}