G\u00fcvenilir Bir Yapay Zeka Ajan\u0131 Olu\u015fturma: Java, Quarkus, LangChain4j ile \u00dcretimde (2. B\u00f6l\u00fcm: Bellek)<\/p>\n
Merhaba! G\u00fcvenilir bir yapay zeka ajan\u0131 olu\u015fturma serimizin ikinci b\u00f6l\u00fcm\u00fcne ho\u015f geldiniz. \u0130lk b\u00f6l\u00fcmde, Java, Quarkus ve LangChain4j’yi kullanarak bir yapay zeka ajan\u0131 nas\u0131l olu\u015fturulaca\u011f\u0131n\u0131 g\u00f6sterdik. Bu b\u00f6l\u00fcmde, yapay zeka ajan\u0131n\u0131z\u0131n daha ak\u0131ll\u0131 ve kullan\u0131\u015fl\u0131 olmas\u0131na olanak tan\u0131yan \u00f6nemli bir bile\u015fen olan belle\u011fi ele alaca\u011f\u0131z.<\/p>\n
Bir yapay zeka ajan\u0131, etkile\u015fimli bir sistemdir. Yani, bir kullan\u0131c\u0131yla veya ba\u015fka bir sistemle etkile\u015fime girer ve tepkiler verir. Bu etkile\u015fimler s\u0131ras\u0131nda, ajan \u00f6nemli bilgiler toplar. \u00d6rne\u011fin, bir m\u00fc\u015fteri hizmetleri ajan\u0131, bir kullan\u0131c\u0131n\u0131n \u00f6nceki sorular\u0131n\u0131 ve yan\u0131tlar\u0131n\u0131 hat\u0131rlamal\u0131, b\u00f6ylece daha iyi bir hizmet sunabilir. Bir web taray\u0131c\u0131s\u0131, bir kullan\u0131c\u0131n\u0131n gezinme ge\u00e7mi\u015fini ve tercih etti\u011fi web sitelerini hat\u0131rlamal\u0131, b\u00f6ylece daha ki\u015fiselle\u015ftirilmi\u015f bir deneyim sunabilir.<\/p>\n
Bir yapay zeka ajan\u0131, daha \u00f6nceki etkile\u015fimlerden elde edilen bilgileri hat\u0131rlayabildi\u011fi zaman, daha ak\u0131ll\u0131, daha kullan\u0131\u015fl\u0131 ve daha etkili hale gelir. Bu, ajanlar\u0131n daha iyi tahminlerde bulunmas\u0131n\u0131, daha alakal\u0131 yan\u0131tlar vermesini ve daha ki\u015fiselle\u015ftirilmi\u015f deneyimler sunmas\u0131n\u0131 sa\u011flar.<\/p>\n
LangChain4j, yapay zeka ajanlar\u0131n\u0131n belle\u011fini uygulamak i\u00e7in \u00e7e\u015fitli ara\u00e7lar sunar. En yayg\u0131n kullan\u0131lan y\u00f6ntemlerden biri, *Memory* arabirimini kullanmakt\u0131r.<\/p>\n
**Memory** arabirimi, bir yapay zeka ajan\u0131 taraf\u0131ndan toplanan bilgileri depolamak ve geri almak i\u00e7in bir mekanizma sa\u011flar. LangChain4j, farkl\u0131 t\u00fcrde bellekleri destekler. Bunlardan baz\u0131lar\u0131 \u015funlard\u0131r:<\/p>\n
* **Chat History Memory:** Bir sohbet ge\u00e7mi\u015fini saklamak i\u00e7in kullan\u0131l\u0131r.
\n* **Vector Store Memory:** Bilgiyi vekt\u00f6rler halinde depolamak i\u00e7in kullan\u0131l\u0131r. Bu, benzer i\u00e7eri\u011fi bulmak i\u00e7in kullan\u0131\u015fl\u0131d\u0131r.
\n* **Conversation Buffer Memory:** Son birka\u00e7 sohbet d\u00f6n\u00fc\u015f\u00fcn\u00fc saklamak i\u00e7in kullan\u0131l\u0131r.<\/p>\n
\u0130\u015fte LangChain4j’yi kullanarak bir sohbet ge\u00e7mi\u015fi belle\u011fi nas\u0131l uygulanaca\u011f\u0131n\u0131 g\u00f6steren bir \u00f6rnek:<\/p>\n
“`java
\nimport io.quarkus.runtime.StartupEvent;
\nimport io.quarkus.runtime.annotations.Startup;
\nimport java.util.concurrent.atomic.AtomicInteger;
\nimport javax.inject.Inject;
\nimport javax.inject.Singleton;
\nimport org.langchain.chains.ConversationalRetrievalQAChain;
\nimport org.langchain.chains.RetrievalQAChain;
\nimport org.langchain.chains.llm.LLMChain;
\nimport org.langchain.chains.llm.LLMChain.LLMChainBuilder;
\nimport org.langchain.embeddings.HuggingFaceEmbeddings;
\nimport org.langchain.embeddings.OpenAIEmbeddings;
\nimport org.langchain.memory.ChatHistoryMemory;
\nimport org.langchain.memory.ConversationBufferMemory;
\nimport org.langchain.memory.Memory;
\nimport org.langchain.memory.VectorStoreMemory;
\nimport org.langchain.model.ChatOpenAI;
\nimport org.langchain.model.TextEmbeddings;
\nimport org.langchain.retrievers.base.CacheRetriever;
\nimport org.langchain.retrievers.base.Retriever;
\nimport org.langchain.retrievers.vectorstore.VectorStoreRetriever;
\nimport org.langchain.schema.LanguageModel;
\nimport org.langchain.text_splitter.RecursiveCharacterTextSplitter;
\nimport org.langchain.vectorstores.Chroma;
\nimport org.langchain.vectorstores.InMemoryVectorStore;
\nimport org.langchain.vectorstores.VectorStore;<\/p>\n
@Startup
\n@Singleton
\npublic class Agent {
\n @Inject
\n LLMChain llmChain;<\/p>\n
public void start(StartupEvent event) {
\n ChatOpenAI chatOpenAI = ChatOpenAI.builder().modelName(“gpt-3.5-turbo”).build();
\n \/\/ TextEmbeddings embeddings = new OpenAIEmbeddings();
\n TextEmbeddings embeddings = new HuggingFaceEmbeddings(“sentence-transformers\/all-mpnet-base-v2”);<\/p>\n
\/\/ Create a memory to store the chat history.
\n Memory memory = new ChatHistoryMemory();
\n \/\/ Memory memory = new ConversationBufferMemory();<\/p>\n
\/\/ Create a retriever that retrieves relevant information from the memory.
\n Retriever retriever = new CacheRetriever(
\n new VectorStoreRetriever(
\n new VectorStoreMemory(memory, embeddings))
\n );<\/p>\n
\/\/ Create a chain that uses the memory and retriever to answer questions.
\n RetrievalQAChain chain = ConversationalRetrievalQAChain
\n .builder()
\n .retriever(retriever)
\n .llmChain(llmChain)
\n .build();
\n System.out.println(“Yapay Zeka Ajan\u0131 ba\u015flat\u0131ld\u0131.”);
\n }
\n}
\n“`<\/p>\n
Bu kod par\u00e7as\u0131, bir sohbet ge\u00e7mi\u015fi belle\u011fi (`ChatHistoryMemory`) ve bir vekt\u00f6r deposu geri al\u0131c\u0131 (`VectorStoreRetriever`) olu\u015fturur. Geri al\u0131c\u0131, bellekteki ilgili bilgileri al\u0131r. Son olarak, bellek ve geri al\u0131c\u0131y\u0131 kullanan ve sorular\u0131 yan\u0131tlayan bir zincir (`ConversationalRetrievalQAChain`) olu\u015fturulur.<\/p>\n
Yapay zeka ajanlar\u0131nda bellek kullanman\u0131n bir\u00e7ok avantaj\u0131 vard\u0131r:<\/p>\n
* **Daha iyi tahminler:** Bellek, ajanlar\u0131n ge\u00e7mi\u015f etkile\u015fimlerden \u00f6\u011frenmesini ve gelecekte daha iyi tahminlerde bulunmas\u0131n\u0131 sa\u011flar.
\n* **Daha alakal\u0131 yan\u0131tlar:** Bellek, ajanlar\u0131n kullan\u0131c\u0131 ba\u011flam\u0131n\u0131 anlamas\u0131n\u0131 ve daha alakal\u0131 yan\u0131tlar vermesini sa\u011flar.
\n* **Daha ki\u015fiselle\u015ftirilmi\u015f deneyimler:** Bellek, ajanlar\u0131n kullan\u0131c\u0131 tercihlerini ve ge\u00e7mi\u015f etkile\u015fimlerini hat\u0131rlamas\u0131n\u0131 ve daha ki\u015fiselle\u015ftirilmi\u015f deneyimler sunmas\u0131n\u0131 sa\u011flar.<\/p>\n
Bellek, g\u00fcvenilir bir yapay zeka ajan\u0131 olu\u015fturman\u0131n \u00f6nemli bir par\u00e7as\u0131d\u0131r. LangChain4j, \u00e7e\u015fitli bellek t\u00fcrlerini uygulamak i\u00e7in ara\u00e7lar sunar. Do\u011fru bellek t\u00fcr\u00fcn\u00fc se\u00e7mek, ajan\u0131n \u00f6zel ihtiya\u00e7lar\u0131na ve kullan\u0131m durumuna ba\u011fl\u0131d\u0131r.<\/p>\n
Bir sonraki b\u00f6l\u00fcmde, yapay zeka ajanlar\u0131nda bellek kullan\u0131m\u0131yla ilgili birka\u00e7 ger\u00e7ek d\u00fcnya \u00f6rne\u011fini inceleyece\u011fiz. Bu \u00f6rnekler, farkl\u0131 bellek t\u00fcrlerinin nas\u0131l kullan\u0131labilece\u011fini ve yapay zeka ajanlar\u0131n\u0131 daha ak\u0131ll\u0131 ve kullan\u0131\u015fl\u0131 hale getirdi\u011fini g\u00f6sterecektir.<\/p>\n
Bir yapay zeka ajan\u0131 i\u00e7in bellek se\u00e7erken a\u015fa\u011f\u0131daki ipu\u00e7lar\u0131n\u0131 g\u00f6z \u00f6n\u00fcnde bulundurun:<\/p>\n
* **Kullan\u0131m durumunuzu g\u00f6z \u00f6n\u00fcnde bulundurun:** Hangi t\u00fcr bilgiler saklanmal\u0131? Ne kadar uzun s\u00fcre saklanmal\u0131?
\n* **Verimlilik:** Belle\u011fin performans\u0131 nas\u0131l? Bilgiye h\u0131zl\u0131 ve verimli bir \u015fekilde eri\u015filebilir mi?
\n* **G\u00fcvenlik:** Bellek, hassas bilgileri g\u00fcvenli bir \u015fekilde sakl\u0131yor mu?<\/p>\n
* Fatih Soysal Blog<\/a> #Etiketler
\n* LangChain Belgeleri<\/a>
\n* Hugging Face Embeddings<\/a><\/p>\n
\n#YapayZeka #Ajan #Bellek #LangChain4j #Java #Quarkus #\u00dcretim #ChatHistoryMemory #VectorStoreMemory #ConversationBufferMemory<\/p>\n","protected":false},"excerpt":{"rendered":"G\u00fcvenilir Bir Yapay Zeka Ajan\u0131 Olu\u015fturma: Java, Quarkus, LangChain4j ile \u00dcretimde (2. B\u00f6l\u00fcm: Bellek) Merhaba! G\u00fcvenilir bir yapay…","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"csco_page_header_type":"","csco_page_load_nextpost":"","csco_page_subscribe_form":"","csco_page_contact_form":"","footnotes":""},"categories":[1],"tags":[],"class_list":{"0":"post-3728","1":"post","2":"type-post","3":"status-publish","4":"format-standard","6":"category-genel","7":"cs-entry","8":"cs-video-wrap"},"yoast_head":"\n