{
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"id": "3b314ab2-3c38-4b8d-9b30-7be726e2bd04",
"metadata": {},
"outputs": [],
"source": [
"import os\n",
"from dotenv import load_dotenv, find_dotenv\n",
"_ = load_dotenv(find_dotenv())\n",
"openai_api_key = os.environ[\"OPENAI_API_KEY\"]"
]
},
{
"cell_type": "markdown",
"id": "b4ea0ec9-8f65-40ce-8c5a-701d140f49d3",
"metadata": {},
"source": [
"## Basic app: a very simple Agent that will decide which external plugin to use"
]
},
{
"cell_type": "code",
"execution_count": 2,
"id": "2dec1e74-2f8a-4917-bbbb-bfaa3e75dc51",
"metadata": {},
"outputs": [],
"source": [
"from langchain_openai import OpenAI"
]
},
{
"cell_type": "code",
"execution_count": 3,
"id": "b7ad8794-04e5-4477-a43b-dc3387d2d830",
"metadata": {},
"outputs": [],
"source": [
"llm = OpenAI()"
]
},
{
"cell_type": "markdown",
"id": "7b48ffd6-2365-407e-97e9-5bf68a6e022a",
"metadata": {},
"source": [
"**Load Google Search module and credentials**"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "89e84730-7dba-4227-82bd-0e209c57e92e",
"metadata": {},
"outputs": [],
"source": [
"from langchain.utilities import GoogleSearchAPIWrapper"
]
},
{
"cell_type": "markdown",
"id": "65ed25d2-06e5-4894-b34f-e35f7179034c",
"metadata": {},
"source": [
"You can sign up for both of these keys at the urls below:\n",
"\n",
"[GOOGLE_API_KEY](https://console.cloud.google.com/apis/credentials)\n",
"[GOOGLE_CSE_ID](https://programmablesearchengine.google.com/controlpanel/create)"
]
},
{
"cell_type": "code",
"execution_count": 5,
"id": "0bdb4485-3ae6-4126-b144-b550b367e1af",
"metadata": {},
"outputs": [],
"source": [
"os.environ[\"GOOGLE_API_KEY\"] = os.environ[\"GOOGLE_API_KEY\"]"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "7801c694-9fd7-4982-bfb3-ae00d0e4dd32",
"metadata": {},
"outputs": [],
"source": [
"os.environ[\"GOOGLE_CSE_ID\"] = os.environ[\"GOOGLE_CSE_ID\"]"
]
},
{
"cell_type": "code",
"execution_count": 7,
"id": "f412229b-85ab-420d-9e44-083cc2301036",
"metadata": {},
"outputs": [],
"source": [
"#!pip install google-api-python-client"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "5ac53d18-f52b-4139-a60d-d9cccb3eb7ae",
"metadata": {},
"outputs": [],
"source": [
"search = GoogleSearchAPIWrapper()"
]
},
{
"cell_type": "markdown",
"id": "cd91704e-9968-4f11-b781-4b3675f83077",
"metadata": {},
"source": [
"**Load module to connect with external URLs**"
]
},
{
"cell_type": "code",
"execution_count": 9,
"id": "cb13f037-9430-41a5-87fc-2b366bde2c1a",
"metadata": {},
"outputs": [],
"source": [
"from langchain.utilities import TextRequestsWrapper"
]
},
{
"cell_type": "code",
"execution_count": 10,
"id": "a6a45a1a-5a1c-4d79-908f-258fa3f83072",
"metadata": {},
"outputs": [],
"source": [
"requests = TextRequestsWrapper()"
]
},
{
"cell_type": "markdown",
"id": "002c10ab-015e-4d95-acb7-76ef8581c069",
"metadata": {},
"source": [
"**Set the tools that will use our agent**"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "f466bbf8-72b1-4849-89e7-c1fe0ca76fde",
"metadata": {},
"outputs": [],
"source": [
"from langchain.agents import Tool"
]
},
{
"cell_type": "code",
"execution_count": 12,
"id": "8cf2982c-6f56-417c-9eb5-ea049e4d7af4",
"metadata": {},
"outputs": [],
"source": [
"tools = [\n",
" Tool(\n",
" name=\"google-search\",\n",
" func=search.run,\n",
" description=\"\"\"\n",
" useful when you need to search google\n",
" to answer questions about current events\n",
" \"\"\"\n",
" ),\n",
" Tool(\n",
" name=\"request\",\n",
" func=requests.get,\n",
" description=\"\"\"\n",
" useful when you need to make a request to a URL\n",
" \"\"\"\n",
" )\n",
"]"
]
},
{
"cell_type": "markdown",
"id": "0766918d-7e8c-410a-86cd-258082b52be5",
"metadata": {},
"source": [
"**Agent initialization and configuration**"
]
},
{
"cell_type": "code",
"execution_count": 13,
"id": "24e569d8-85a0-492e-aae8-5f1b777e2e0b",
"metadata": {},
"outputs": [],
"source": [
"from langchain.agents import initialize_agent"
]
},
{
"cell_type": "code",
"execution_count": 14,
"id": "fe51fa19-329b-4d6f-8b93-60a796fddb15",
"metadata": {},
"outputs": [],
"source": [
"from langchain.agents import AgentType"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "819a6ffa-9d6a-4b90-8a59-fd9a48d693dc",
"metadata": {},
"outputs": [],
"source": [
"from langchain.agents import load_tools"
]
},
{
"cell_type": "code",
"execution_count": 16,
"id": "0358117c-709f-4637-8451-dfd4369723c0",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/Users/juliocolomer/.pyenv/versions/3.11.4/envs/venv020124/lib/python3.11/site-packages/langchain_core/_api/deprecation.py:117: LangChainDeprecationWarning: The function `initialize_agent` was deprecated in LangChain 0.1.0 and will be removed in 0.2.0. Use Use new agent constructor methods like create_react_agent, create_json_agent, create_structured_chat_agent, etc. instead.\n",
" warn_deprecated(\n"
]
}
],
"source": [
"my_agent = initialize_agent(\n",
" tools=tools,\n",
" llm=llm,\n",
" agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION,\n",
" verbose=True,\n",
" max_iterations=3\n",
")"
]
},
{
"cell_type": "markdown",
"id": "da302337-c07c-416d-8868-8db34cd65c0a",
"metadata": {},
"source": [
"**Ask questions to the agent and see how she decides which tool to use**"
]
},
{
"cell_type": "markdown",
"id": "fa09f26d-1037-404b-8e57-48a665d0a622",
"metadata": {},
"source": [
"*Careful: operations with agents are more expensive than regular operations with OpenAI*"
]
},
{
"cell_type": "code",
"execution_count": 17,
"id": "2ac904ae-b978-4104-be3d-ff0e4367ed51",
"metadata": {},
"outputs": [],
"source": [
"question = \"Which team won the 2010 soccer world chapionship?\""
]
},
{
"cell_type": "code",
"execution_count": 18,
"id": "221229b4-a50b-404b-9265-926f8f66e5cb",
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"/Users/juliocolomer/.pyenv/versions/3.11.4/envs/venv020124/lib/python3.11/site-packages/langchain_core/_api/deprecation.py:117: LangChainDeprecationWarning: The function `__call__` was deprecated in LangChain 0.1.0 and will be removed in 0.2.0. Use invoke instead.\n",
" warn_deprecated(\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n",
"\n",
"\u001b[1m> Entering new AgentExecutor chain...\u001b[0m\n",
"\u001b[32;1m\u001b[1;3m We should search google to find the answer\n",
"Action: google-search\n",
"Action Input: 2010 soccer world championship winner\u001b[0m\n",
"Observation: \u001b[36;1m\u001b[1;3mSpain became the eighth nation to win the tournament and the first European nation to win a World Cup hosted outside its home continent: all previous World Cups ... We're sorry! We are experiencing a problem with our website right now. We are working to correct this issue. Continue on fifa.com/inside. ^. Cookie Settings ... The win was Spain's first World Cup title, and it lifted them to first place in the FIFA World Rankings with the Netherlands moving into second. Spain also ... See our comprehensive FIFA Men's World Cup history guide for everything you need about the tournament. FIFA Men's World Cup results and which countries have ... For example, the winner of the Oceanian zone and the fifth-placed team from the Asian zone entered a play-off for a spot in the 2010 World Cup. From the 1938 ... May 15, 2004 ... South Africa has won the right to host the 2010 soccer World Cup – the first to be held on the African continent. European champions Spain beat Cristiano Ronaldo's Portugal in the second round and Paraguay in the quarter-finals, both 1-0. They saw off Germany by the same ... Nov 14, 2022 ... 2010 Spain – Spain's possession style of football saw them win their first ever World Cup in South Africa in 2010. 2014 Germany – After making ... Sep 30, 2019 ... South Africa wins the 2010 bid to host the Soccer World Cup ... On 15 May 2004, the president of the Federation of International Football ... 2010 FIFA World Cup ; 3,178,856 (49,670 per match) · Uruguay Diego Forlán · Germany Thomas Müller · Netherlands Wesley Sneijder · Spain David Villa (5 goals each).\u001b[0m\n",
"Thought:\u001b[32;1m\u001b[1;3m I now know the final answer\n",
"Final Answer: Spain\u001b[0m\n",
"\n",
"\u001b[1m> Finished chain.\u001b[0m\n"
]
}
],
"source": [
"response = my_agent(question)"
]
},
{
"cell_type": "code",
"execution_count": 19,
"id": "5695e4d9-fa9a-46c4-acec-933d7471effe",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Spain\n"
]
}
],
"source": [
"print(response[\"output\"])"
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "b8aafc7b-4480-4490-9b33-512321bc9c89",
"metadata": {},
"outputs": [],
"source": [
"question2 = {\n",
" \"input\": \"Tell me about what subject are the comments in this webpage https://news.ycombinator.com/item?id=34425779\"\n",
"}"
]
},
{
"cell_type": "code",
"execution_count": 21,
"id": "4f2513d6-36d8-49e6-9afb-5e087d5e703b",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"\n",
"\n",
"\u001b[1m> Entering new AgentExecutor chain...\u001b[0m\n",
"\u001b[32;1m\u001b[1;3m Comments are typically written by users to express their opinions or share information about a particular subject.\n",
"Action: request\n",
"Action Input: https://news.ycombinator.com/item?id=34425779\u001b[0m\n",
"Observation: \u001b[33;1m\u001b[1;3m\n",
" \n",
" I mean *deep* embeddings (i.e., sequences of hidden states, the ones are compute... | Hacker News
\n",
" I mean deep embeddings (i.e., sequences of hidden states, the ones are computed by all those interactions) , not the shallow embeddings of token ids in the first layer of the model! Those deep embeddings are much richer representations.
Imagine if you and others building apps had access to "GPT3 deep sequence embeddings v1.0" via an API.
\n",
" Not quite. My understanding is that OpenAI's various embeddings APIs return only a single vector per document, instead of the sequence of hidden states corresponding to each predicted next token in the response generated by a GPT-type LLM.
Imagine getting generated text from a GPT LLM that comes with a deep embedding of each generated token's "contextual meaning":
\n",
" \n",
" \n",
"\u001b[0m\n",
"Thought:\u001b[32;1m\u001b[1;3m I need to find and read the comments on the webpage.\n",
"Action: request\n",
"Action Input: https://news.ycombinator.com/item?id=34425779\u001b[0m\n",
"Observation: \u001b[33;1m\u001b[1;3m\n",
" \n",
" I mean *deep* embeddings (i.e., sequences of hidden states, the ones are compute... | Hacker News
\n",
" I mean deep embeddings (i.e., sequences of hidden states, the ones are computed by all those interactions) , not the shallow embeddings of token ids in the first layer of the model! Those deep embeddings are much richer representations.
Imagine if you and others building apps had access to "GPT3 deep sequence embeddings v1.0" via an API.
\n",
" Not quite. My understanding is that OpenAI's various embeddings APIs return only a single vector per document, instead of the sequence of hidden states corresponding to each predicted next token in the response generated by a GPT-type LLM.
Imagine getting generated text from a GPT LLM that comes with a deep embedding of each generated token's "contextual meaning":
![](\"s.gif\")
![](\"y18.svg\")
Imagine if you and others building apps had access to "GPT3 deep sequence embeddings v1.0" via an API.
\n", "