mirror of
https://github.com/hwchase17/langchain
synced 2024-10-29 17:07:25 +00:00
282 lines
11 KiB
Plaintext
282 lines
11 KiB
Plaintext
{
|
|
"cells": [
|
|
{
|
|
"attachments": {},
|
|
"cell_type": "markdown",
|
|
"id": "9e9b7651",
|
|
"metadata": {},
|
|
"source": [
|
|
"# How to use a SmartLLMChain\n",
|
|
"\n",
|
|
"A SmartLLMChain is a form of self-critique chain that can help you if have particularly complex questions to answer. Instead of doing a single LLM pass, it instead performs these 3 steps:\n",
|
|
"1. Ideation: Pass the user prompt n times through the LLM to get n output proposals (called \"ideas\"), where n is a parameter you can set \n",
|
|
"2. Critique: The LLM critiques all ideas to find possible flaws and picks the best one \n",
|
|
"3. Resolve: The LLM tries to improve upon the best idea (as chosen in the critique step) and outputs it. This is then the final output.\n",
|
|
"\n",
|
|
"SmartLLMChains are based on the SmartGPT workflow proposed in https://youtu.be/wVzuvf9D9BU.\n",
|
|
"\n",
|
|
"Note that SmartLLMChains\n",
|
|
"- use more LLM passes (ie n+2 instead of just 1)\n",
|
|
"- only work then the underlying LLM has the capability for reflection, which smaller models often don't\n",
|
|
"- only work with underlying models that return exactly 1 output, not multiple\n",
|
|
"\n",
|
|
"This notebook demonstrates how to use a SmartLLMChain."
|
|
]
|
|
},
|
|
{
|
|
"attachments": {},
|
|
"cell_type": "markdown",
|
|
"id": "714dede0",
|
|
"metadata": {},
|
|
"source": [
|
|
"##### Same LLM for all steps"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 1,
|
|
"id": "d3f7fb22",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"import os\n",
|
|
"\n",
|
|
"os.environ[\"OPENAI_API_KEY\"] = \"...\""
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 2,
|
|
"id": "10e5ece6",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"from langchain.prompts import PromptTemplate\n",
|
|
"from langchain.chat_models import ChatOpenAI\n",
|
|
"from langchain_experimental.smart_llm import SmartLLMChain"
|
|
]
|
|
},
|
|
{
|
|
"attachments": {},
|
|
"cell_type": "markdown",
|
|
"id": "1780da51",
|
|
"metadata": {},
|
|
"source": [
|
|
"As example question, we will use \"I have a 12 liter jug and a 6 liter jug. I want to measure 6 liters. How do I do it?\". This is an example from the original SmartGPT video (https://youtu.be/wVzuvf9D9BU?t=384). While this seems like a very easy question, LLMs struggle do these kinds of questions that involve numbers and physical reasoning.\n",
|
|
"\n",
|
|
"As we will see, all 3 initial ideas are completely wrong - even though we're using GPT4! Only when using self-reflection do we get a correct answer. "
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "054af6b1",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"hard_question = \"I have a 12 liter jug and a 6 liter jug. I want to measure 6 liters. How do I do it?\""
|
|
]
|
|
},
|
|
{
|
|
"attachments": {},
|
|
"cell_type": "markdown",
|
|
"id": "8049cecd",
|
|
"metadata": {},
|
|
"source": [
|
|
"So, we first create an LLM and prompt template"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 3,
|
|
"id": "811ea8e1",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"prompt = PromptTemplate.from_template(hard_question)\n",
|
|
"llm = ChatOpenAI(temperature=0, model_name=\"gpt-4\")"
|
|
]
|
|
},
|
|
{
|
|
"attachments": {},
|
|
"cell_type": "markdown",
|
|
"id": "50b602e4",
|
|
"metadata": {},
|
|
"source": [
|
|
"Now we can create a SmartLLMChain"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 4,
|
|
"id": "8cd49199",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"chain = SmartLLMChain(llm=llm, prompt=prompt, n_ideas=3, verbose=True)"
|
|
]
|
|
},
|
|
{
|
|
"attachments": {},
|
|
"cell_type": "markdown",
|
|
"id": "6a72f276",
|
|
"metadata": {},
|
|
"source": [
|
|
"Now we can use the SmartLLM as a drop-in replacement for our LLM. E.g.:"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 5,
|
|
"id": "074e5e75",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"\n",
|
|
"\n",
|
|
"\u001b[1m> Entering new SmartLLMChain chain...\u001b[0m\n",
|
|
"Prompt after formatting:\n",
|
|
"\u001b[32;1m\u001b[1;3mI have a 12 liter jug and a 6 liter jug. I want to measure 6 liters. How do I do it?\u001b[0m\n",
|
|
"Idea 1:\n",
|
|
"\u001b[36;1m\u001b[1;3m1. Fill the 6-liter jug completely.\n",
|
|
"2. Pour the water from the 6-liter jug into the 12-liter jug.\n",
|
|
"3. Fill the 6-liter jug again.\n",
|
|
"4. Carefully pour the water from the 6-liter jug into the 12-liter jug until the 12-liter jug is full.\n",
|
|
"5. The amount of water left in the 6-liter jug will be exactly 6 liters.\u001b[0m\n",
|
|
"Idea 2:\n",
|
|
"\u001b[36;1m\u001b[1;3m1. Fill the 6-liter jug completely.\n",
|
|
"2. Pour the water from the 6-liter jug into the 12-liter jug.\n",
|
|
"3. Fill the 6-liter jug again.\n",
|
|
"4. Carefully pour the water from the 6-liter jug into the 12-liter jug until the 12-liter jug is full.\n",
|
|
"5. Since the 12-liter jug is now full, there will be 2 liters of water left in the 6-liter jug.\n",
|
|
"6. Empty the 12-liter jug.\n",
|
|
"7. Pour the 2 liters of water from the 6-liter jug into the 12-liter jug.\n",
|
|
"8. Fill the 6-liter jug completely again.\n",
|
|
"9. Pour the water from the 6-liter jug into the 12-liter jug, which already has 2 liters in it.\n",
|
|
"10. Now, the 12-liter jug will have exactly 6 liters of water (2 liters from before + 4 liters from the 6-liter jug).\u001b[0m\n",
|
|
"Idea 3:\n",
|
|
"\u001b[36;1m\u001b[1;3m1. Fill the 6-liter jug completely.\n",
|
|
"2. Pour the water from the 6-liter jug into the 12-liter jug.\n",
|
|
"3. Fill the 6-liter jug again.\n",
|
|
"4. Carefully pour the water from the 6-liter jug into the 12-liter jug until the 12-liter jug is full.\n",
|
|
"5. The amount of water left in the 6-liter jug will be exactly 6 liters.\u001b[0m\n",
|
|
"Critique:\n",
|
|
"\u001b[33;1m\u001b[1;3mIdea 1:\n",
|
|
"1. Fill the 6-liter jug completely. (No flaw)\n",
|
|
"2. Pour the water from the 6-liter jug into the 12-liter jug. (No flaw)\n",
|
|
"3. Fill the 6-liter jug again. (No flaw)\n",
|
|
"4. Carefully pour the water from the 6-liter jug into the 12-liter jug until the 12-liter jug is full. (Flaw: The 12-liter jug will never be full in this step, as it can hold 12 liters and we are only pouring 6 liters into it.)\n",
|
|
"5. The amount of water left in the 6-liter jug will be exactly 6 liters. (Flaw: This statement is incorrect, as there will be no water left in the 6-liter jug after pouring it into the 12-liter jug.)\n",
|
|
"\n",
|
|
"Idea 2:\n",
|
|
"1. Fill the 6-liter jug completely. (No flaw)\n",
|
|
"2. Pour the water from the 6-liter jug into the 12-liter jug. (No flaw)\n",
|
|
"3. Fill the 6-liter jug again. (No flaw)\n",
|
|
"4. Carefully pour the water from the 6-liter jug into the 12-liter jug until the 12-liter jug is full. (Flaw: The 12-liter jug will never be full in this step, as it can hold 12 liters and we are only pouring 6 liters into it.)\n",
|
|
"5. Since the 12-liter jug is now full, there will be 2 liters of water left in the 6-liter jug. (Flaw: This statement is incorrect, as the 12-liter jug will not be full and there will be no water left in the 6-liter jug.)\n",
|
|
"6. Empty the 12-liter jug. (No flaw)\n",
|
|
"7. Pour the 2 liters of water from the 6-liter jug into the 12-liter jug. (Flaw: This step is based on the incorrect assumption that there are 2 liters of water left in the 6-liter jug.)\n",
|
|
"8. Fill the 6-liter jug completely again. (No flaw)\n",
|
|
"9. Pour the water from the 6-liter jug into the 12-liter jug, which already has 2 liters in it. (Flaw: This step is based on the incorrect assumption that there are 2 liters of water in the 12-liter jug.)\n",
|
|
"10. Now, the 12-liter jug will have exactly 6 liters of water (2 liters from before + 4 liters from the 6-liter jug). (Flaw: This conclusion is based on the incorrect assumptions made in the previous steps.)\n",
|
|
"\n",
|
|
"Idea 3:\n",
|
|
"1. Fill the 6-liter jug completely. (No flaw)\n",
|
|
"2. Pour the water from the 6-liter jug into the 12-liter jug. (No flaw)\n",
|
|
"3. Fill the 6-liter jug again. (No flaw)\n",
|
|
"4. Carefully pour the water from the 6-liter jug into the 12-liter jug until the 12-liter jug is full. (Flaw: The 12-liter jug will never be full in this step, as it can hold 12 liters and we are only pouring 6 liters into it.)\n",
|
|
"5. The amount of water left in the 6-liter jug will be exactly 6 liters. (Flaw: This statement is incorrect, as there will be no water left in the 6-liter jug after pouring it into the 12-liter jug.)\u001b[0m\n",
|
|
"Resolution:\n",
|
|
"\u001b[32;1m\u001b[1;3m1. Fill the 12-liter jug completely.\n",
|
|
"2. Pour the water from the 12-liter jug into the 6-liter jug until the 6-liter jug is full.\n",
|
|
"3. The amount of water left in the 12-liter jug will be exactly 6 liters.\u001b[0m\n",
|
|
"\n",
|
|
"\u001b[1m> Finished chain.\u001b[0m\n"
|
|
]
|
|
},
|
|
{
|
|
"data": {
|
|
"text/plain": [
|
|
"'1. Fill the 12-liter jug completely.\\n2. Pour the water from the 12-liter jug into the 6-liter jug until the 6-liter jug is full.\\n3. The amount of water left in the 12-liter jug will be exactly 6 liters.'"
|
|
]
|
|
},
|
|
"execution_count": 5,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"chain.run({})"
|
|
]
|
|
},
|
|
{
|
|
"attachments": {},
|
|
"cell_type": "markdown",
|
|
"id": "bbfebea1",
|
|
"metadata": {},
|
|
"source": [
|
|
"##### Different LLM for different steps"
|
|
]
|
|
},
|
|
{
|
|
"attachments": {},
|
|
"cell_type": "markdown",
|
|
"id": "5be6ec08",
|
|
"metadata": {},
|
|
"source": [
|
|
"You can also use different LLMs for the different steps by passing `ideation_llm`, `critique_llm` and `resolve_llm`. You might want to do this to use a more creative (i.e., high-temperature) model for ideation and a more strict (i.e., low-temperature) model for critique and resolution."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 8,
|
|
"id": "9c33fa19",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"chain = SmartLLMChain(\n",
|
|
" ideation_llm=ChatOpenAI(temperature=0.9, model_name=\"gpt-4\"),\n",
|
|
" llm=ChatOpenAI(\n",
|
|
" temperature=0, model_name=\"gpt-4\"\n",
|
|
" ), # will be used for critique and resolution as no specific llms are given\n",
|
|
" prompt=prompt,\n",
|
|
" n_ideas=3,\n",
|
|
" verbose=True,\n",
|
|
")"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"id": "886c1cc1",
|
|
"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": "3.9.1"
|
|
}
|
|
},
|
|
"nbformat": 4,
|
|
"nbformat_minor": 5
|
|
}
|