■ VectorStoreRetriever 클래스의 configurable_fields 메소드에서 search_kwargs 인자를 사용해 RunnableConfigurableFields 객체를 만드는 방법을 보여준다. ※ PINECONE_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶
■ VectorStoreRetriever 클래스의 invoke 메소드를 사용해 벡터 유사도 검색하는 방법을 보여준다. ※ PINECONE_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶ main.py
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 |
from dotenv import load_dotenv from langchain_openai import OpenAIEmbeddings from langchain_pinecone import PineconeVectorStore load_dotenv() openAIEmbeddings = OpenAIEmbeddings() pineconeVectorStore = PineconeVectorStore(index_name = "test-example", embedding = openAIEmbeddings) # Pinecone 사이트에서 test-example 인덱스가 존재해야 한다. pineconeVectorStore.add_texts(["i worked at kensho" ], namespace = "harrison") pineconeVectorStore.add_texts(["i worked at facebook"], namespace = "ankush" ) vectorStoreRetriever1 = pineconeVectorStore.as_retriever(search_kwargs = {"namespace" : "ankush"}) responseDocumentList1 = vectorStoreRetriever1.invoke("where did i work?") print(responseDocumentList1) print("-" * 50) vectorStoreRetriever2 = pineconeVectorStore.as_retriever(search_kwargs = {"namespace" : "harrison"}) responseDocumentList2 = vectorStoreRetriever2.invoke("where did i work?") print(responseDocumentList2) print("-" * 50) """ [ Document(id = 'a7e34e14-fe06-439e-aa0b-d40fd46412c2', metadata = {}, page_content = 'i worked at facebook'), Document(id = 'fa276204-878a-41e1-b85a-6f081d8a0036', metadata = {}, page_content = 'i worked at facebook') ] -------------------------------------------------- [ Document(id = '14d1d34b-e0a0-4259-9bc3-173c8538b830', metadata = {}, page_content = 'i worked at kensho'), Document(id = '4b73972a-0b28-487f-a801-d00c81527928', metadata = {}, page_content = 'i worked at kensho') ] -------------------------------------------------- """ |
■ PineconeVectorStore 클래스의 as_retriever 메소드에서 search_kwargs 인자를 사용해 VectorStoreRetriever 객체를 만드는 방법을 보여준다. ※ PINECONE_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶
■ PineconeVectorStore 클래스의 add_texts 메소드에서 namespace 인자를 사용해 텍스트를 추가하는 방법을 보여준다. ※ PINECONE_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶ main.py
■ langchain-pinecone 패키지를 설치하는 방법을 보여준다. 1. 명령 프롬프트를 실행한다. 2. 명령 프롬프트에서 아래 명령을 실행한다. ▶ 실행 명령
|
1 2 3 |
pip install langchain-pinecone |
■ PineconeVectorStore 클래스의 생성자에서 index_name/embedding 인자를 사용해 PineconeVectorStore 객체를 만드는 방법을 보여준다. ※ PINECONE_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶ main.py
■ 답변 생성 후 다음 모델에서 자체 답변에 인용문을 주석으로 추가하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶
■ EmbeddingsFilter 클래스의 compress_documents 메소드를 사용해 검색 문서 후처리를 통해 컨텐츠를 압축하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env 파일에 정의한다.
■ ChatPromptTemplate 클래스에서 프롬프트 문자열을 직접 사용해 모델 응답 구조에서 문서 ID와 인묭문을 구하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env
■ ChatOpenAI 클래스의 with_structured_output 메소드를 사용해 모델 응답 구조에서 문서 ID를 인용하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env 파일에 정의한다.
■ WikipediaRetriever 클래스를 사용해 위키피디아 검색하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶ main.py
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 |
from dotenv import load_dotenv from langchain_community.retrievers import WikipediaRetriever from typing import List from langchain_core.documents import Document from langchain_core.prompts import ChatPromptTemplate from langchain_openai import ChatOpenAI from langchain_core.runnables import RunnablePassthrough from langchain_core.output_parsers import StrOutputParser load_dotenv() wikipediaRetriever = WikipediaRetriever(top_k_results = 6, doc_content_chars_max = 2000) runnableSequence1 = (lambda x: x["input"]) | wikipediaRetriever def getStringDocumentList(documentList: List[Document]): return "\n\n".join(document.page_content for document in documentList) chatPromptTemplate = ChatPromptTemplate.from_messages( [ ("system", "You're a helpful AI assistant. Given a user question and some Wikipedia article snippets, answer the user question. If none of the articles answer the question, just say you don't know.\n\nHere are the Wikipedia articles : {context}"), ("human", "{input}") ] ) chatOpenAI = ChatOpenAI(model = "gpt-4o-mini") runnableSequence2 = ( RunnablePassthrough.assign(context = (lambda x : getStringDocumentList(x["context"]))) | chatPromptTemplate | chatOpenAI | StrOutputParser() ) runnableSequence3 = RunnablePassthrough.assign(context = runnableSequence1).assign(answer = runnableSequence2) responseDictionary = runnableSequence3.invoke({"input" : "How fast are cheetahs?"}) print(responseDictionary) """ { 'input' : 'How fast are cheetahs?', 'context' : [ Document( metadata = { 'title' : 'Cheetah', 'summary' : 'The cheetah (Acinonyx jubatus) is a large cat and the fastest land animal. It has a tawny to creamy white or pale buff fur that is marked with evenly spaced, solid black spots. The head is small and rounded, with a short snout and black tear-like facial streaks. It reaches 67–94 cm (26–37 in) at the shoulder, and the head-and-body length is between 1.1 and 1.5 m (3 ft 7 in and 4 ft 11 in). Adults weigh between 21 and 72 kg (46 and 159 lb). The cheetah is capable of running at 93 to 104 km/h (58 to 65 mph); it has evolved specialized adaptations for speed, including a light build, long thin legs and a long tail.\nThe cheetah was first described in the late 18th century. Four subspecies are recognised today that are native to Africa and central Iran. An African subspecies was introduced to India in 2022. It is now distributed mainly in small, fragmented populations in northwestern, eastern and southern Africa and central Iran. It lives in a variety of habitats such as savannahs in the Serengeti, arid mountain ranges in the Sahara, and hilly desert terrain.\nThe cheetah lives in three main social groups: females and their cubs, male "coalitions", and solitary males. While females lead a nomadic life searching for prey in large home ranges, males are more sedentary and instead establish much smaller territories in areas with plentiful prey and access to females. The cheetah is active during the day, with peaks during dawn and dusk. It feeds on small- to medium-sized prey, mostly weighing under 40 kg (88 lb), and prefers medium-sized ungulates such as impala, springbok and Thomson\'s gazelles. The cheetah typically stalks its prey within 60–100 m (200–330 ft) before charging towards it, trips it during the chase and bites its throat to suffocate it to death. It breeds throughout the year. After a gestation of nearly three months, females give birth to a litter of three or four cubs. Cheetah cubs are highly vulnerable to predation by other large carnivores. They are weaned at around four months and are independent by around 20 months of age.\nThe cheetah is threatened by habitat loss, conflict with humans, poaching and high susceptibility to diseases. The global cheetah population was estimated in 2021 at 6,517; it is listed as Vulnerable on the IUCN Red List. It has been widely depicted in art, literature, advertising, and animation. It was tamed in ancient Egypt and trained for hunting ungulates in the Arabian Peninsula and India. It has been kept in zoos since the early 19th century.', 'source' : 'https://en.wikipedia.org/wiki/Cheetah' }, page_content = 'The cheetah (Acinonyx jubatus) is a large cat and the fastest land animal. It has a tawny to creamy white or pale buff fur that is marked with evenly spaced, solid black spots. The head is small and rounded, with a short snout and black tear-like facial streaks. It reaches 67–94 cm (26–37 in) at the shoulder, and the head-and-body length is between 1.1 and 1.5 m (3 ft 7 in and 4 ft 11 in). Adults weigh between 21 and 72 kg (46 and 159 lb). The cheetah is capable of running at 93 to 104 km/h (58 to 65 mph); it has evolved specialized adaptations for speed, including a light build, long thin legs and a long tail.\nThe cheetah was first described in the late 18th century. Four subspecies are recognised today that are native to Africa and central Iran. An African subspecies was introduced to India in 2022. It is now distributed mainly in small, fragmented populations in northwestern, eastern and southern Africa and central Iran. It lives in a variety of habitats such as savannahs in the Serengeti, arid mountain ranges in the Sahara, and hilly desert terrain.\nThe cheetah lives in three main social groups: females and their cubs, male "coalitions", and solitary males. While females lead a nomadic life searching for prey in large home ranges, males are more sedentary and instead establish much smaller territories in areas with plentiful prey and access to females. The cheetah is active during the day, with peaks during dawn and dusk. It feeds on small- to medium-sized prey, mostly weighing under 40 kg (88 lb), and prefers medium-sized ungulates such as impala, springbok and Thomson\'s gazelles. The cheetah typically stalks its prey within 60–100 m (200–330 ft) before charging towards it, trips it during the chase and bites its throat to suffocate it to death. It breeds throughout the year. After a gestation of nearly three months, females give birth to a litter of three or four cubs. Cheetah cubs are highly vulnerable to predation by other large carnivores. They are weaned a' ), Document ( metadata = { 'title' : 'Southeast African cheetah', 'summary' : 'The Southeast African cheetah (Acinonyx jubatus jubatus) is the nominate cheetah subspecies native to East and Southern Africa. The Southern African cheetah lives mainly in the lowland areas and deserts of the Kalahari, the savannahs of Okavango Delta, and the grasslands of the Transvaal region in South Africa. In Namibia, cheetahs are mostly found in farmlands. In India, four cheetahs of the subspecies are living in Kuno National Park in Madhya Pradesh after having been introduced there.', 'source' : 'https://en.wikipedia.org/wiki/Southeast_African_cheetah' }, page_content = 'The Southeast African cheetah (Acinonyx jubatus jubatus) is the nominate cheetah subspecies native to East and Southern Africa. The Southern African cheetah lives mainly in the lowland areas and deserts of the Kalahari, the savannahs of Okavango Delta, and the grasslands of the Transvaal region in South Africa. In Namibia, cheetahs are mostly found in farmlands. In India, four cheetahs of the subspecies are living in Kuno National Park in Madhya Pradesh after having been introduced there.\n\n\n== Taxonomy ==\n\nThe Southern African cheetah was first described by German naturalist Johann Christian Daniel von Schreber in his book Die Säugethiere in Abbildungen nach der Natur mit Beschreibungen (The Mammals illustrated as in Nature with Descriptions), published in 1775. Schreber described the species on basis of a specimen from the Cape of Good Hope. It is therefore the nominate subspecies. Subpopulations have been called "South African cheetah" and "Namibian cheetah."\nFollowing Schreber\'s description, other naturalists and zoologists also described cheetah specimens from many parts of Southern and East Africa that today are all considered synonyms of A. j. jubatus:\n\nFelis guttata proposed in 1804 by Johann Hermann;\nFelis fearonii proposed in 1834 by Andrew Smith;\nFelis lanea proposed in 1877 by Philip Sclater;\nAcinonyx jubatus obergi proposed in 1913 by Max Hilzheimer;\nAcinonyx jubatus ngorongorensis proposed in 1913 by Hilzheimer on basis of a specimen from Ngorongoro, German East Africa;\nAcinonyx jubatus velox proposed in 1913 by Edmund Heller on basis of a cheetah that was shot by Kermit Roosevelt in June 1909 in the Kenyan highlands.\nAcinonyx rex proposed in 1927 by Reginald Innes Pocock on basis of a specimen from the Umvukwe Range in Rhodesia.\nIn 2005, the authors of Mammal Species of the World grouped A. j. guttata, A. j. lanea, A. j. obergi, and A. j. rex under A j. jubatus, whilst recognizing A. j. raineyi and A. j. velox as valid taxa and considering P. l. ngoron' ), Document( metadata = { 'title' : 'Fastest animals', 'summary' : 'This is a list of the fastest animals in the world, by types of animal.', 'source': 'https://en.wikipedia.org/wiki/Fastest_animals' }, page_content = "This is a list of the fastest animals in the world, by types of animal.\n\n\n== Fastest organism ==\nThe peregrine falcon is the fastest bird, and the fastest member of the animal kingdom, with a diving speed of over 300 km/h (190 mph). The fastest land animal is the cheetah. Among the fastest animals in the sea is the black marlin, with uncertain and conflicting reports of recorded speeds.\nWhen drawing comparisons between different classes of animals, an alternative unit is sometimes used for organisms: body length per second. On this basis the 'fastest' organism on earth, relative to its body length, is the Southern Californian mite, Paratarsotomus macropalpis, which has a speed of 322 body lengths per second. The equivalent speed for a human, running as fast as this mite, would be 1,300 mph (2,092 km/h), or approximately Mach 1.7. The speed of the P. macropalpis is far in excess of the previous record holder, the Australian tiger beetle Rivacindela hudsoni, which is the fastest insect in the world relative to body size, with a recorded speed of 1.86 metres per second (6.7 km/h; 4.2 mph), or 171 body lengths per second. The cheetah, the fastest land mammal, scores at only 16 body lengths per second.\n\n\n== Invertebrates ==\n\n\n== Fish ==\nDue to physical constraints, fish may be incapable of exceeding swim speeds of 36 km/h (22 mph). The larger reported figures below are therefore highly questionable:\n\n\n== Amphibians ==\n\n\n== Reptiles ==\n\n\n== Birds ==\n\n\n== Mammals ==\n\n\n== See also ==\nSpeed record\n\n\n== Notes ==\n\n\n== References ==" ), Document( metadata = { 'title' : 'Footspeed', 'summary' : 'Footspeed, or sprint speed, is the maximum speed at which a human can run. It is affected by many factors, varies greatly throughout the population, and is important in athletics and many sports, such as association football, Australian rules football, American football, track and field, field hockey, tennis, baseball, and basketball.', 'source' : 'https://en.wikipedia.org/wiki/Footspeed' }, page_content = 'Footspeed, or sprint speed, is the maximum speed at which a human can run. It is affected by many factors, varies greatly throughout the population, and is important in athletics and many sports, such as association football, Australian rules football, American football, track and field, field hockey, tennis, baseball, and basketball.\n\n\n== Factors in speed ==\nThe key determinant of footspeed in sprinting is the predominance of one distinct type of muscle fibre over another, specifically the ratio of fast-twitch muscles to slow-twitch muscles in a sprinter\'s physical makeup. Though fast-twitch muscles produce no more energy than slow-twitch muscles when they contract, they do so more rapidly through a process of anaerobic metabolism, though at the cost of inferior efficiency over longer periods of firing. The average human has an almost-equal ratio of fast-twitch to slow-twitch fibers, but top sprinters may have as much as 80% fast-twitch fibers, while top long-distance runners may have only 20%. This ratio is believed to have genetic origins, though some assert that it can be adjusted by muscle training. "Speed camps" and "Speed Training Manuals", which purport to provide fractional increases in maximum footspeed, are popular among budding professional athletes, and some sources estimate that 17–19% of speed can be trained.\nThough good running form is useful in increasing speed, fast and slow runners have been shown to move their legs at nearly the same rate – it is the force exerted by the leg on the ground that separates fast sprinters from slow. Top short-distance runners exert as much as four times their body weight in pressure on the running surface. For this reason, muscle mass in the legs, relative to total body weight, is a key factor in maximizing footspeed.\n\n\n== Limits of speed ==\nThe record is 44.72 km/h (27.78 mph), measured between meter 60 and meter 80 of the 100 meters sprint at the 2009 World Championships in Athletics by Usain Bolt. (Bolt\'s ave' ), Document( metadata = { 'title' : 'Pursuit predation', 'summary' : "Pursuit predation is a form of predation in which predators actively give chase to their prey, either solitarily or as a group. It is an alternate predation strategy to ambush predation — pursuit predators rely on superior speed, endurance and/or teamwork to seize the prey, while ambush predators use concealment, luring, exploiting of surroundings and the element of surprise to capture the prey. While the two patterns of predation are not mutually exclusive, morphological differences in an organism's body plan can create an evolutionary bias favoring either type of predation.\nPursuit predation is typically observed in carnivorous species within the kingdom Animalia, such as cheetahs, lions, wolves and early Homo species. The chase can be initiated either by the predator, or by the prey if it is alerted to a predator's presence and attempt to flee before the predator gets close. The chase ends either when the predator successfully catches up and tackles the prey, or when the predator abandons the attempt after the prey outruns it and escapes.\nOne particular form of pursuit predation is persistence hunting, where the predator stalks the prey slowly but persistently to wear it down physically with fatigue or overheating; some animals are examples of both types of pursuit.\n\n", 'source' : 'https://en.wikipedia.org/wiki/Pursuit_predation' }, page_content = "Pursuit predation is a form of predation in which predators actively give chase to their prey, either solitarily or as a group. It is an alternate predation strategy to ambush predation — pursuit predators rely on superior speed, endurance and/or teamwork to seize the prey, while ambush predators use concealment, luring, exploiting of surroundings and the element of surprise to capture the prey. While the two patterns of predation are not mutually exclusive, morphological differences in an organism's body plan can create an evolutionary bias favoring either type of predation.\nPursuit predation is typically observed in carnivorous species within the kingdom Animalia, such as cheetahs, lions, wolves and early Homo species. The chase can be initiated either by the predator, or by the prey if it is alerted to a predator's presence and attempt to flee before the predator gets close. The chase ends either when the predator successfully catches up and tackles the prey, or when the predator abandons the attempt after the prey outruns it and escapes.\nOne particular form of pursuit predation is persistence hunting, where the predator stalks the prey slowly but persistently to wear it down physically with fatigue or overheating; some animals are examples of both types of pursuit.\n\n\n== Strategy ==\nThere is still uncertainty as to whether predators behave with a general tactic or strategy while preying. However, among pursuit predators there are several common behaviors. Often, predators will scout potential prey, assessing prey quantity and density prior to engaging in a pursuit. Certain predators choose to pursue prey primarily in a group of conspecifics; these animals are known as pack hunters or group pursuers. Other species choose to hunt alone. These two behaviors are typically due to differences in hunting success, where some groups are very successful in groups and others are more successful alone. Pursuit predators may also choose to either exhaust their metabolic r" ), Document( metadata = { 'title' : 'Gepard-class fast attack craft', 'summary' : 'The Type 143A Gepard class was a class of missile bearing fast attack craft (German: Schnellboot) and the last one in service with the German Navy before the remaining four operational ships were decommissioned on 16 November 2016. The Ghana Navy operates two such ships. \nIt is an evolution of the Albatros class, the main difference being the replacement of the second 76 mm gun by the RAM system. The Gepard-class vessels were gradually supplemented by Braunschweig-class corvettes and later replaced completely by them.\nThe ships in class were named after small to medium-sized predatory animals; Gepard is German for "cheetah".', 'source' : 'https://en.wikipedia.org/wiki/Gepard-class_fast_attack_craft' }, page_content = 'The Type 143A Gepard class was a class of missile bearing fast attack craft (German: Schnellboot) and the last one in service with the German Navy before the remaining four operational ships were decommissioned on 16 November 2016. The Ghana Navy operates two such ships. \nIt is an evolution of the Albatros class, the main difference being the replacement of the second 76 mm gun by the RAM system. The Gepard-class vessels were gradually supplemented by Braunschweig-class corvettes and later replaced completely by them.\nThe ships in class were named after small to medium-sized predatory animals; Gepard is German for "cheetah".\n\n\n== List of ships ==\n\nThe "S" and the number are part of the ship\'s full name. When the ships were first commissioned, their designation included only the number; however, the crews petitioned for full names, and the decision was made to combine the original names with the additional animal name.\nSince 1 July 2006, all ships had formed part of the 7. Schnellbootgeschwader (7th Fast Patrol Boat Squadron), whereas for the eight years prior the flotilla was split into (hulls S 76–S 80) 2. Schnellbootgeschwader (2nd Fast Patrol Boat Squadron), and (hulls S 71–S 75) 7. Schnellbootgeschwader. The squadron was stationed in Warnemünde, where both predecessor squadrons had been based.\n\n\n== Gallery ==\n\n\t\t\n\t\t\t\n\t\t\t\n\t\t\n\t\t\n\t\t\t\n\t\t\t\n\t\t\n\t\t\n\t\t\t\n\t\t\t\n\t\t\n\t\t\n\t\t\t\n\t\t\t\n\t\t\n\n\n== Notes ==\n\n\n== References ==\n\n"Schnellboot GEPARD-Klasse" (in German). German Navy. Retrieved 10 July 2013.' ) ], 'answer' : 'Cheetahs can run at speeds of 93 to 104 km/h (58 to 65 mph), making them the fastest land animals.' } """ |
▶ requirements.txt
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 |
aiohappyeyeballs==2.4.4 aiohttp==3.11.9 aiosignal==1.3.1 annotated-types==0.7.0 anyio==4.6.2.post1 attrs==24.2.0 beautifulsoup4==4.12.3 certifi==2024.8.30 charset-normalizer==3.4.0 colorama==0.4.6 dataclasses-json==0.6.7 distro==1.9.0 frozenlist==1.5.0 greenlet==3.1.1 h11==0.14.0 httpcore==1.0.7 httpx==0.28.0 httpx-sse==0.4.0 idna==3.10 jiter==0.8.0 jsonpatch==1.33 jsonpointer==3.0.0 langchain==0.3.9 langchain-community==0.3.9 langchain-core==0.3.21 langchain-openai==0.2.10 langchain-text-splitters==0.3.2 langsmith==0.1.147 marshmallow==3.23.1 multidict==6.1.0 mypy-extensions==1.0.0 numpy==2.1.3 openai==1.56.0 orjson==3.10.12 packaging==24.2 propcache==0.2.1 pydantic==2.10.2 pydantic-settings==2.6.1 pydantic_core==2.27.1 python-dotenv==1.0.1 PyYAML==6.0.2 regex==2024.11.6 requests==2.32.3 requests-toolbelt==1.0.0 sniffio==1.3.1 soupsieve==2.6 SQLAlchemy==2.0.36 tenacity==9.0.0 tiktoken==0.8.0 tqdm==4.67.1 typing-inspect==0.9.0 typing_extensions==4.12.2 urllib3==2.2.3 wikipedia==1.4.0 yarl==1.18.3 |
■ ChatOpenAI 클래스의 with_structured_output 메소드를 사용해 RAG 애플리케이션에서 모델 응답 구조의 소스를 구하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env 파일에
■ LCEL을 사용해 RAG 애플리케이션에서 소스를 구하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶ main.py
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 |
import bs4 from dotenv import load_dotenv from langchain_community.document_loaders import WebBaseLoader from langchain_text_splitters import RecursiveCharacterTextSplitter from langchain_chroma import Chroma from langchain_openai import OpenAIEmbeddings from langchain_openai import ChatOpenAI from langchain_core.prompts import ChatPromptTemplate from langchain_core.output_parsers import StrOutputParser from langchain_core.runnables import RunnablePassthrough load_dotenv() webBaseLoader = WebBaseLoader( web_paths = ("https://lilianweng.github.io/posts/2023-06-23-agent/",), bs_kwargs = dict(parse_only = bs4.SoupStrainer(class_ = ("post-content", "post-title", "post-header"))) ) documentList = webBaseLoader.load() recursiveCharacterTextSplitter = RecursiveCharacterTextSplitter(chunk_size = 1000, chunk_overlap = 200) splitDocumentList = recursiveCharacterTextSplitter.split_documents(documentList) chroma = Chroma.from_documents(documents=splitDocumentList, embedding=OpenAIEmbeddings()) vectorStoreRetriever = chroma.as_retriever() runnableSequence1 = (lambda x : x["input"]) | vectorStoreRetriever def formatDocumentList(documentList): return "\n\n".join(document.page_content for document in documentList) chatPromptTemplate = ChatPromptTemplate.from_messages( [ ("system", "You are an assistant for question-answering tasks. Use the following pieces of retrieved context to answer the question. If you don't know the answer, say that you don't know. Use three sentences maximum and keep the answer concise.\n\n{context}"), ("human", "{input}") ] ) chatOpenAI = ChatOpenAI(model = "gpt-4o-mini") runnableSequence2 = ( { "input" : lambda x : x["input"], # input query "context" : lambda x : formatDocumentList(x["context"]), # context } | chatPromptTemplate | chatOpenAI | StrOutputParser() ) runnableSequence3 = RunnablePassthrough.assign(context = runnableSequence1).assign(answer = runnableSequence2) responseDictionary = runnableSequence3.invoke({"input" : "What is Task Decomposition"}) print(responseDictionary) """ { 'input' : 'What is Task Decomposition', 'context' : [ Document( metadata = {'source' : 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content = 'Fig. 1. Overview of a LLM-powered autonomous agent system.\nComponent One: Planning#\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\nTask Decomposition#\nChain of thought (CoT; Wei et al. 2022) has become a standard prompting technique for enhancing model performance on complex tasks. The model is instructed to “think step by step” to utilize more test-time computation to decompose hard tasks into smaller and simpler steps. CoT transforms big tasks into multiple manageable tasks and shed lights into an interpretation of the model’s thinking process.' ), Document( metadata = {'source' : 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content = 'Tree of Thoughts (Yao et al. 2023) extends CoT by exploring multiple reasoning possibilities at each step. It first decomposes the problem into multiple thought steps and generates multiple thoughts per step, creating a tree structure. The search process can be BFS (breadth-first search) or DFS (depth-first search) with each state evaluated by a classifier (via a prompt) or majority vote.\nTask decomposition can be done (1) by LLM with simple prompting like "Steps for XYZ.\\n1.", "What are the subgoals for achieving XYZ?", (2) by using task-specific instructions; e.g. "Write a story outline." for writing a novel, or (3) with human inputs.' ), Document( metadata = {'source' : 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content = 'The AI assistant can parse user input to several tasks: [{"task": task, "id", task_id, "dep": dependency_task_ids, "args": {"text": text, "image": URL, "audio": URL, "video": URL}}]. The "dep" field denotes the id of the previous task which generates a new resource that the current task relies on. A special tag "-task_id" refers to the generated text image, audio and video in the dependency task with id as task_id. The task MUST be selected from the following options: {{ Available Task List }}. There is a logical relationship between tasks, please note their order. If the user input can\'t be parsed, you need to reply empty JSON. Here are several cases for your reference: {{ Demonstrations }}. The chat history is recorded as {{ Chat History }}. From this chat history, you can find the path of the user-mentioned resources for your task planning.' ), Document( metadata = {'source' : 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content = 'Fig. 11. Illustration of how HuggingGPT works. (Image source: Shen et al. 2023)\nThe system comprises of 4 stages:\n(1) Task planning: LLM works as the brain and parses the user requests into multiple tasks. There are four attributes associated with each task: task type, ID, dependencies, and arguments. They use few-shot examples to guide LLM to do task parsing and planning.\nInstruction:' ) ], 'answer' : 'Task Decomposition is the process of breaking down a complicated task into smaller, more manageable steps. Techniques like Chain of Thought (CoT) and Tree of Thoughts (ToT) are used to enhance model performance by allowing the model to think step by step and explore multiple reasoning possibilities. This approach aids in planning and execution by transforming large tasks into simpler, interpretable components.' } """ |
▶ requirements.txt
■ create_retrieval_chain 함수를 사용해 RAG 애플리케이션에서 소스를 구하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶ main.py
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
import bs4 from dotenv import load_dotenv from langchain_community.document_loaders import WebBaseLoader from langchain_text_splitters import RecursiveCharacterTextSplitter from langchain_chroma import Chroma from langchain_openai import OpenAIEmbeddings from langchain_openai import ChatOpenAI from langchain_core.prompts import ChatPromptTemplate from langchain.chains.combine_documents import create_stuff_documents_chain from langchain.chains import create_retrieval_chain load_dotenv() webBaseLoader = WebBaseLoader( web_paths = ("https://lilianweng.github.io/posts/2023-06-23-agent/",), bs_kwargs = dict(parse_only = bs4.SoupStrainer(class_ = ("post-content", "post-title", "post-header"))) ) documentList = webBaseLoader.load() recursiveCharacterTextSplitter = RecursiveCharacterTextSplitter(chunk_size = 1000, chunk_overlap = 200) splitDocumentList = recursiveCharacterTextSplitter.split_documents(documentList) chroma = Chroma.from_documents(documents=splitDocumentList, embedding=OpenAIEmbeddings()) vectorStoreRetriever = chroma.as_retriever() chatOpenAI = ChatOpenAI(model = "gpt-4o-mini") chatPromptTemplate = ChatPromptTemplate.from_messages( [ ("system", "You are an assistant for question-answering tasks. Use the following pieces of retrieved context to answer the question. If you don't know the answer, say that you don't know. Use three sentences maximum and keep the answer concise.\n\n{context}"), ("human", "{input}") ] ) runnableBinding1 = create_stuff_documents_chain(chatOpenAI, chatPromptTemplate) runnableBinding2 = create_retrieval_chain(vectorStoreRetriever, runnableBinding1) responseDictionary = runnableBinding2.invoke({"input" : "What is Task Decomposition?"}) print(responseDictionary) """ { 'input' : 'What is Task Decomposition?', 'context' : [ Document( metadata = {'source' : 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content = 'Fig. 1. Overview of a LLM-powered autonomous agent system.\nComponent One: Planning#\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\nTask Decomposition#\nChain of thought (CoT; Wei et al. 2022) has become a standard prompting technique for enhancing model performance on complex tasks. The model is instructed to “think step by step” to utilize more test-time computation to decompose hard tasks into smaller and simpler steps. CoT transforms big tasks into multiple manageable tasks and shed lights into an interpretation of the model’s thinking pree structure. The search process can be BFS (breadth-first search) or DFS (depth-first search) with each state evaluated by a classifier (via a prompt) or majority vote.\nTask decomposition can be done (1) by LLM with simple prompting like "Steps for XYZ.\\n1.", "What are the subgoals for achieving XYZ?", (2) by using task-specific instructions; e.g. "Write a story outline." for writing a novel, or (3) with human inputs.' ), Document( metadata = {'source' : 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content = 'Resources:\n1. Internet access for searches and information gathering.\n2. Long Term memory management.\n3. GPT-3.5 powered Agents for delegation of simple tasks.\n4. File output.\n\nPerformance Evaluation:\n1. Continuously review and analyze your actions to ensure you are performing to the best of your abilities.\n2. Constructively self-criticize your big-picture behavior constantly.\n3. Reflect on past decisions and strategies to refine your approach.\n4. Every command has a cost, so be smart and efficient. Aim to complete tasks in the least number of steps.' ), Document( metadata = {'source' : 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content = "(3) Task execution: Expert models execute on the specific tasks and log results.\nInstruction:\n\nWith the input and the inference results, the AI assistant needs to describe the process and results. The previous stages can be formed as - User Input: {{ User Input }}, Task Planning: {{ Tasks }}, Model Selection: {{ Model Assignment }}, Task Execution: {{ Predictions }}. You must first answer the user's request in a straightforward manner. Then describe the task process and show your analysis and model inference results to the user in the first person. If inference results contain a file path, must tell the user the complete file path." ) ], 'answer' : 'Task Decomposition is the process of breaking down a complex task into smaller, manageable steps to facilitate easier execution and understanding. Techniques like Chain of Thought (CoT) and Tree of Thoughts enhance this process by guiding models to think step-by-step and explore multiple reasoning possibilities at each stage. This approach helps in improving performance on complex tasks by simplifying them into subgoals.' } """ |
▶
■ RunnableBinding 클래스의 astream_events 메소드를 사용해 대화형 RAG 체인에서 검색기 전달전 재구성 사용자 질문을 출력하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은
■ RunnableBinding 클래스의 astream 메소드를 사용해 대화형 RAG 체인에서 검색기 전달전 사용자 질문을 재구성하는 방법을 보여준다. ※ 재구성된 질문은 최종 출력의 일부로
■ RunnableBinding 클래스의 pick 메소드를 사용해 RunnableSequence 객체를 만드는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶ main.py
|
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 |
import bs4 from dotenv import load_dotenv from langchain_community.document_loaders import WebBaseLoader from langchain_text_splitters import RecursiveCharacterTextSplitter from langchain_openai import OpenAIEmbeddings from langchain_chroma import Chroma from langchain_openai import ChatOpenAI from langchain_core.prompts import ChatPromptTemplate from langchain.chains.combine_documents import create_stuff_documents_chain from langchain.chains import create_retrieval_chain load_dotenv() webBaseLoader = WebBaseLoader( web_paths = ("https://lilianweng.github.io/posts/2023-06-23-agent/",), bs_kwargs = dict(parse_only = bs4.SoupStrainer(class_ = ("post-content", "post-title", "post-header"))) ) documentList = webBaseLoader.load() recursiveCharacterTextSplitter = RecursiveCharacterTextSplitter(chunk_size = 1000, chunk_overlap = 200) splitDocumentList = recursiveCharacterTextSplitter.split_documents(documentList) chroma = Chroma.from_documents(documents = splitDocumentList, embedding = OpenAIEmbeddings()) vectorStoreRetriever = chroma.as_retriever() chatOpenAI = ChatOpenAI(model = "gpt-4o-mini") chatPromptTemplate = ChatPromptTemplate.from_messages( [ ("system", "You are an assistant for question-answering tasks. Use the following pieces of retrieved context to answer the question. If you don't know the answer, say that you don't know. Use three sentences maximum and keep the answer concise.\n\n{context}"), ("human", "{input}") ] ) runnableBinding1 = create_stuff_documents_chain(chatOpenAI, chatPromptTemplate) # create_retrieval_chain 함수로 구성된 체인은 "input", "context", "answer" 키를 가진 dict 객체를 반환한다. # 여기서 "answer" 키만 토큰 단위로 스트리밍되고, 검색과 같은 다른 구성 요소는 토큰 수준 스트리밍을 지원하지 않는다. runnableBinding2 = create_retrieval_chain(vectorStoreRetriever, runnableBinding1) runnableSequence = runnableBinding2.pick("answer") for addableDict in runnableSequence.stream({"input" : "What is Task Decomposition?"}): print(f"{addableDict}|", end = "", flush = True) print() print("-" * 50) """ |Task| decomposition| is| the| process| of| breaking| down| complex| tasks| into| smaller|,| more| manageable| steps|.| It| often| utilizes| techniques| like| Chain| of| Thought| (|Co|T|)| to| guide| models| in| thinking| step| by| step|,| or| the| Tree| of| Thoughts| approach|,| which| explores| multiple| reasoning| possibilities| at| each| step|.| This| method| enhances| performance| by| allowing| a| clearer| understanding| and| execution| of| complicated| tasks|.|| -------------------------------------------------- """ |
■ RunnableBinding 클래스의 stream 메소드를 사용해 RAG 애플리케이션의 결과를 스트리밍하는 방법을 보여준다. ※ OPENAI_API_KEY 환경 변수 값은 .env 파일에 정의한다. ▶ main.py