mirror of https://github.com/hwchase17/langchain
You cannot select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
428 lines
12 KiB
Plaintext
428 lines
12 KiB
Plaintext
{
|
|
"cells": [
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "a9ab2a39-7c2d-4119-9dc7-8035fdfba3cb",
|
|
"metadata": {},
|
|
"source": [
|
|
"# LangSmith LLM Runs\n",
|
|
"\n",
|
|
"This notebook demonstrates how to directly load data from LangSmith's LLM runs and fine-tune a model on that data.\n",
|
|
"The process is simple and comprises 3 steps.\n",
|
|
"\n",
|
|
"1. Select the LLM runs to train on.\n",
|
|
"2. Use the LangSmithRunChatLoader to load runs as chat sessions.\n",
|
|
"3. Fine-tune your model.\n",
|
|
"\n",
|
|
"Then you can use the fine-tuned model in your LangChain app.\n",
|
|
"\n",
|
|
"Before diving in, let's install our prerequisites.\n",
|
|
"\n",
|
|
"## Prerequisites\n",
|
|
"\n",
|
|
"Ensure you've installed langchain >= 0.0.311 and have configured your environment with your LangSmith API key."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 1,
|
|
"id": "ef488003-514a-48b4-93f1-7de4417abf5d",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"%pip install --upgrade --quiet langchain langchain-openai"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 1,
|
|
"id": "473adce5-c863-49e6-85c3-049e0ec2222e",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"import os\n",
|
|
"import uuid\n",
|
|
"\n",
|
|
"uid = uuid.uuid4().hex[:6]\n",
|
|
"project_name = f\"Run Fine-tuning Walkthrough {uid}\"\n",
|
|
"os.environ[\"LANGCHAIN_TRACING_V2\"] = \"true\"\n",
|
|
"os.environ[\"LANGCHAIN_API_KEY\"] = \"YOUR API KEY\"\n",
|
|
"os.environ[\"LANGCHAIN_PROJECT\"] = project_name"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "8533ab63-d437-492a-aaec-ccca31167bf2",
|
|
"metadata": {},
|
|
"source": [
|
|
"## 1. Select Runs\n",
|
|
"The first step is selecting which runs to fine-tune on. A common case would be to select LLM runs within\n",
|
|
"traces that have received positive user feedback. You can find examples of this in the[LangSmith Cookbook](https://github.com/langchain-ai/langsmith-cookbook/blob/main/exploratory-data-analysis/exporting-llm-runs-and-feedback/llm_run_etl.ipynb) and in the [docs](https://docs.smith.langchain.com/tracing/use-cases/export-runs/local).\n",
|
|
"\n",
|
|
"For the sake of this tutorial, we will generate some runs for you to use here. Let's try fine-tuning a\n",
|
|
"simple function-calling chain."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 2,
|
|
"id": "9a36d27f-2f3b-4148-b94a-9436fe8b00e0",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"from enum import Enum\n",
|
|
"\n",
|
|
"from langchain_core.pydantic_v1 import BaseModel, Field\n",
|
|
"\n",
|
|
"\n",
|
|
"class Operation(Enum):\n",
|
|
" add = \"+\"\n",
|
|
" subtract = \"-\"\n",
|
|
" multiply = \"*\"\n",
|
|
" divide = \"/\"\n",
|
|
"\n",
|
|
"\n",
|
|
"class Calculator(BaseModel):\n",
|
|
" \"\"\"A calculator function\"\"\"\n",
|
|
"\n",
|
|
" num1: float\n",
|
|
" num2: float\n",
|
|
" operation: Operation = Field(..., description=\"+,-,*,/\")\n",
|
|
"\n",
|
|
" def calculate(self):\n",
|
|
" if self.operation == Operation.add:\n",
|
|
" return self.num1 + self.num2\n",
|
|
" elif self.operation == Operation.subtract:\n",
|
|
" return self.num1 - self.num2\n",
|
|
" elif self.operation == Operation.multiply:\n",
|
|
" return self.num1 * self.num2\n",
|
|
" elif self.operation == Operation.divide:\n",
|
|
" if self.num2 != 0:\n",
|
|
" return self.num1 / self.num2\n",
|
|
" else:\n",
|
|
" return \"Cannot divide by zero\""
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 3,
|
|
"id": "89bcc676-27e8-40dc-a4d6-92cf28e0db58",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"{'description': 'A calculator function',\n",
|
|
" 'name': 'Calculator',\n",
|
|
" 'parameters': {'description': 'A calculator function',\n",
|
|
" 'properties': {'num1': {'title': 'Num1', 'type': 'number'},\n",
|
|
" 'num2': {'title': 'Num2', 'type': 'number'},\n",
|
|
" 'operation': {'allOf': [{'description': 'An '\n",
|
|
" 'enumeration.',\n",
|
|
" 'enum': ['+',\n",
|
|
" '-',\n",
|
|
" '*',\n",
|
|
" '/'],\n",
|
|
" 'title': 'Operation'}],\n",
|
|
" 'description': '+,-,*,/'}},\n",
|
|
" 'required': ['num1', 'num2', 'operation'],\n",
|
|
" 'title': 'Calculator',\n",
|
|
" 'type': 'object'}}\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"from pprint import pprint\n",
|
|
"\n",
|
|
"from langchain.utils.openai_functions import convert_pydantic_to_openai_function\n",
|
|
"from langchain_core.pydantic_v1 import BaseModel\n",
|
|
"\n",
|
|
"openai_function_def = convert_pydantic_to_openai_function(Calculator)\n",
|
|
"pprint(openai_function_def)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 4,
|
|
"id": "cd44ff01-22cf-431a-8bf4-29a758d1fcff",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"from langchain.output_parsers.openai_functions import PydanticOutputFunctionsParser\n",
|
|
"from langchain_core.prompts import ChatPromptTemplate\n",
|
|
"from langchain_openai import ChatOpenAI\n",
|
|
"\n",
|
|
"prompt = ChatPromptTemplate.from_messages(\n",
|
|
" [\n",
|
|
" (\"system\", \"You are an accounting assistant.\"),\n",
|
|
" (\"user\", \"{input}\"),\n",
|
|
" ]\n",
|
|
")\n",
|
|
"chain = (\n",
|
|
" prompt\n",
|
|
" | ChatOpenAI().bind(functions=[openai_function_def])\n",
|
|
" | PydanticOutputFunctionsParser(pydantic_schema=Calculator)\n",
|
|
" | (lambda x: x.calculate())\n",
|
|
")"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 5,
|
|
"id": "62da7d8f-5cfc-45a6-946e-2bcda2b0ba1f",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"math_questions = [\n",
|
|
" \"What's 45/9?\",\n",
|
|
" \"What's 81/9?\",\n",
|
|
" \"What's 72/8?\",\n",
|
|
" \"What's 56/7?\",\n",
|
|
" \"What's 36/6?\",\n",
|
|
" \"What's 64/8?\",\n",
|
|
" \"What's 12*6?\",\n",
|
|
" \"What's 8*8?\",\n",
|
|
" \"What's 10*10?\",\n",
|
|
" \"What's 11*11?\",\n",
|
|
" \"What's 13*13?\",\n",
|
|
" \"What's 45+30?\",\n",
|
|
" \"What's 72+28?\",\n",
|
|
" \"What's 56+44?\",\n",
|
|
" \"What's 63+37?\",\n",
|
|
" \"What's 70-35?\",\n",
|
|
" \"What's 60-30?\",\n",
|
|
" \"What's 50-25?\",\n",
|
|
" \"What's 40-20?\",\n",
|
|
" \"What's 30-15?\",\n",
|
|
"]\n",
|
|
"results = chain.batch([{\"input\": q} for q in math_questions], return_exceptions=True)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "cbb1bcae-b922-4d38-b4bd-4b65be400b88",
|
|
"metadata": {},
|
|
"source": [
|
|
"#### Load runs that did not error\n",
|
|
"\n",
|
|
"Now we can select the successful runs to fine-tune on."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 6,
|
|
"id": "d6037992-050d-4ada-a061-860c124f0bf1",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"from langsmith.client import Client\n",
|
|
"\n",
|
|
"client = Client()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 7,
|
|
"id": "0444919a-6f5a-4726-9916-4603b1420d0e",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"successful_traces = {\n",
|
|
" run.trace_id\n",
|
|
" for run in client.list_runs(\n",
|
|
" project_name=project_name,\n",
|
|
" execution_order=1,\n",
|
|
" error=False,\n",
|
|
" )\n",
|
|
"}\n",
|
|
"\n",
|
|
"llm_runs = [\n",
|
|
" run\n",
|
|
" for run in client.list_runs(\n",
|
|
" project_name=project_name,\n",
|
|
" run_type=\"llm\",\n",
|
|
" )\n",
|
|
" if run.trace_id in successful_traces\n",
|
|
"]"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "f365a359-52f7-47ff-8c36-aadc1070b409",
|
|
"metadata": {},
|
|
"source": [
|
|
"## 2. Prepare data\n",
|
|
"Now we can create an instance of LangSmithRunChatLoader and load the chat sessions using its lazy_load() method."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 8,
|
|
"id": "817bc077-c18a-473b-94a4-a7d810d583a8",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"from langchain_community.chat_loaders.langsmith import LangSmithRunChatLoader\n",
|
|
"\n",
|
|
"loader = LangSmithRunChatLoader(runs=llm_runs)\n",
|
|
"\n",
|
|
"chat_sessions = loader.lazy_load()"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "f21a3bbd-1ed4-481b-9640-206b8bf0d751",
|
|
"metadata": {},
|
|
"source": [
|
|
"#### With the chat sessions loaded, convert them into a format suitable for fine-tuning."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 9,
|
|
"id": "9e5ac127-b094-4584-9159-5a6d3d7315c7",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"from langchain_community.adapters.openai import convert_messages_for_finetuning\n",
|
|
"\n",
|
|
"training_data = convert_messages_for_finetuning(chat_sessions)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "188c4978-d85e-4984-a008-a50f6cd6bb84",
|
|
"metadata": {},
|
|
"source": [
|
|
"## 3. Fine-tune the model\n",
|
|
"Now, initiate the fine-tuning process using the OpenAI library."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 10,
|
|
"id": "11d19e28-be49-4801-8065-1a58d13cd192",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"Status=[running]... 349.84s. 17.72s\r"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"import json\n",
|
|
"import time\n",
|
|
"from io import BytesIO\n",
|
|
"\n",
|
|
"import openai\n",
|
|
"\n",
|
|
"my_file = BytesIO()\n",
|
|
"for dialog in training_data:\n",
|
|
" my_file.write((json.dumps({\"messages\": dialog}) + \"\\n\").encode(\"utf-8\"))\n",
|
|
"\n",
|
|
"my_file.seek(0)\n",
|
|
"training_file = openai.files.create(file=my_file, purpose=\"fine-tune\")\n",
|
|
"\n",
|
|
"job = openai.fine_tuning.jobs.create(\n",
|
|
" training_file=training_file.id,\n",
|
|
" model=\"gpt-3.5-turbo\",\n",
|
|
")\n",
|
|
"\n",
|
|
"# Wait for the fine-tuning to complete (this may take some time)\n",
|
|
"status = openai.fine_tuning.jobs.retrieve(job.id).status\n",
|
|
"start_time = time.time()\n",
|
|
"while status != \"succeeded\":\n",
|
|
" print(f\"Status=[{status}]... {time.time() - start_time:.2f}s\", end=\"\\r\", flush=True)\n",
|
|
" time.sleep(5)\n",
|
|
" status = openai.fine_tuning.jobs.retrieve(job.id).status\n",
|
|
"\n",
|
|
"# Now your model is fine-tuned!"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "54c4cead-500d-41dd-8333-2defde634396",
|
|
"metadata": {},
|
|
"source": [
|
|
"## 4. Use in LangChain\n",
|
|
"\n",
|
|
"After fine-tuning, use the resulting model ID with the ChatOpenAI model class in your LangChain app."
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 11,
|
|
"id": "7f45b281-1dfa-43cb-bd28-99fa7e9f45d1",
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"# Get the fine-tuned model ID\n",
|
|
"job = openai.fine_tuning.jobs.retrieve(job.id)\n",
|
|
"model_id = job.fine_tuned_model\n",
|
|
"\n",
|
|
"# Use the fine-tuned model in LangChain\n",
|
|
"from langchain_openai import ChatOpenAI\n",
|
|
"\n",
|
|
"model = ChatOpenAI(\n",
|
|
" model=model_id,\n",
|
|
" temperature=1,\n",
|
|
")"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 12,
|
|
"id": "7d3b5845-6385-42d1-9f7d-5ea798dc2cd9",
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"data": {
|
|
"text/plain": [
|
|
"AIMessage(content='Let me calculate that for you.')"
|
|
]
|
|
},
|
|
"execution_count": 12,
|
|
"metadata": {},
|
|
"output_type": "execute_result"
|
|
}
|
|
],
|
|
"source": [
|
|
"(prompt | model).invoke({\"input\": \"What's 56/7?\"})"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"id": "5b8c2c79-ce27-4f37-b1b2-5977db8c4e84",
|
|
"metadata": {},
|
|
"source": [
|
|
"Now you have successfully fine-tuned a model using data from LangSmith LLM runs!"
|
|
]
|
|
}
|
|
],
|
|
"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": "3.11.5"
|
|
}
|
|
},
|
|
"nbformat": 4,
|
|
"nbformat_minor": 5
|
|
}
|