Microsoft MVP성태의 닷넷 이야기
.NET Framework: 2125. C# - Semantic Kernel의 Semantic Memory 사용 예제 [링크 복사], [링크+제목 복사],
조회: 11926
글쓴 사람
정성태 (techsharer at outlook.com)
홈페이지
첨부 파일

(시리즈 글이 8개 있습니다.)
.NET Framework: 2117. C# - (OpenAI 기반의) Microsoft Semantic Kernel을 이용한 자연어 처리
; https://www.sysnet.pe.kr/2/0/13345

.NET Framework: 2118. C# - Semantic Kernel의 Prompt chaining 예제
; https://www.sysnet.pe.kr/2/0/13347

.NET Framework: 2119. C# - Semantic Kernel의 "Basic Loading of the Kernel" 예제
; https://www.sysnet.pe.kr/2/0/13348

.NET Framework: 2120. C# - Semantic Kernel의 Skill과 Function 사용 예제
; https://www.sysnet.pe.kr/2/0/13349

.NET Framework: 2121. C# - Semantic Kernel의 대화 문맥 유지
; https://www.sysnet.pe.kr/2/0/13352

.NET Framework: 2123. C# - Semantic Kernel의 ChatGPT 대화 구현
; https://www.sysnet.pe.kr/2/0/13355

.NET Framework: 2124. C# - Semantic Kernel의 Planner 사용 예제
; https://www.sysnet.pe.kr/2/0/13357

.NET Framework: 2125. C# - Semantic Kernel의 Semantic Memory 사용 예제
; https://www.sysnet.pe.kr/2/0/13358




C# - Semantic Kernel의 Semantic Memory 사용 예제

ASK, Kernel, Planner, Skill을 지나 이제 Memory 예제까지 왔습니다. ^^

samples/notebooks/dotnet/06-memory-and-embeddings.ipynb
; https://github.com/microsoft/semantic-kernel/blob/main/samples/notebooks/dotnet/06-memory-and-embeddings.ipynb

Memory는 간단하게 설명하면 전체적인 문맥을 기억하기 위한 수단입니다. 문맥이라고 하니, 지난 Context 예제가 있었는데요,

C# - Semantic Kernel의 대화 문맥 유지
; https://www.sysnet.pe.kr/2/0/13352

사실 위에서 구현한 ContextVariables는 그다지 특별한 것이 아니고, 지금까지의 대화 내용을 전부 누적시켜 설정하는 것에 지나지 않습니다. 일례로, 이전 예제의 코드를 보면,

var context = new ContextVariables();
var history = "";
context.Set("history", history);

...[생략]...

var bot_answer = await kernel.RunAsync(context, skill["Chat"]);
Console.WriteLine(bot_answer);

history += $"\nHuman: {human_input}\nAssistant: {bot_answer}\n";
context.Update(history);

history 문자열에 대화의 내용을 전부 합쳐 Update 시킨 후 RunAsync에 전달하는 식입니다. 여기서 문제는, 이런 식으로 누적시켜 대화를 이어나가면 결국 Model에서 정한 Token 제약에 금방 걸리고 맙니다.

바로 이런 문제를 해결하기 위해, 단기/장기로 대화 상태를 저장할 수 있는 메모리가 필요한 것입니다. 이 외에도, 메모리를 이용하면 또 다른 활용이 가능합니다. 예를 들어, 사내에서 개발한 제품의 매뉴얼이 있다고 가정해 보겠습니다. 물론, OpenAI가 크롤링한 웹 데이터에 해당 매뉴얼이 있다면 상관없겠지만 그렇지 않을 수도 있습니다. 그런 상황이라면 그 매뉴얼 전체를 Chat 내에서 인식시키려 할 때 또다시 Token 제약에 걸립니다.

이런 문제를 Memory를 이용해 해결할 수 있는데요, 이를 위해서는 2가지 구성요소(Memory Storage와 Embedding backend)를 Kernel에 추가해야 합니다. 방법은 예제 코드에 잘 나옵니다. ^^

우선 Embedding backend는 KernelConfig을 이용해 새롭게 "text-embedding-ada-002" 모델을 추가합니다.

KernelConfig kernelConfig = new KernelConfig();
kernelConfig.AddOpenAITextCompletionService("default", "text-davinci-003", apiKey);
kernelConfig.AddOpenAITextEmbeddingGenerationService("ada", "text-embedding-ada-002", apiKey);

그다음 Memory Storage는, Application이 실행 중일 때만 보관이 되도록 VolatileMemoryStore를 사용합니다.

var kernel = Kernel.Builder
    .WithConfiguration(kernelConfig)
    .WithMemoryStorage(new VolatileMemoryStore())
    .Build();

이 외에도, Azure Cosmos DB, PostgreSQL, SQLite 등으로 유지할 수 있다고 합니다. nuget을 찾아보면 관련 타입들이 아마도 다음의 패키지들에 포함된 것 같습니다.

Microsoft.SemanticKernel.Connectors.Memory.Sqlite
; https://www.nuget.org/packages/Microsoft.SemanticKernel.Connectors.Memory.Sqlite

Microsoft.SemanticKernel.Connectors.Memory.Qdrant 
; https://www.nuget.org/packages/Microsoft.SemanticKernel.Connectors.Memory.Qdrant

Microsoft.SemanticKernel.Connectors.Memory.AzureCognitiveSearch
; https://www.nuget.org/packages/Microsoft.SemanticKernel.Connectors.Memory.AzureCognitiveSearch

Microsoft.SemanticKernel.Connectors.Memory.Postgres
; https://github.com/microsoft/semantic-kernel/tree/main/dotnet/src/Connectors/Connectors.Memory.Postgres

당연히 모두 현재(2023-05-17)는 Preview 버전이고 PostgreSQL의 경우에는 찾을 수 없었습니다.

자, 그럼 이제 남은 것은 Memory Store에 문맥 정보를 추가하는 일이 남았는데요, 이를 위해 SaveInformationAsync 메서드를 사용할 수 있습니다.

const string MemoryCollectionName = "aboutMe";

await kernel.Memory.SaveInformationAsync(MemoryCollectionName, id: "info1", text: "My name is Andrea");
await kernel.Memory.SaveInformationAsync(MemoryCollectionName, id: "info2", text: "I currently work as a tourist operator");
await kernel.Memory.SaveInformationAsync(MemoryCollectionName, id: "info3", text: "I currently live in Seattle and have been living there since 2005");
await kernel.Memory.SaveInformationAsync(MemoryCollectionName, id: "info4", text: "I visited France and Italy five times since 2015");
await kernel.Memory.SaveInformationAsync(MemoryCollectionName, id: "info5", text: "My family is from New York");

여기서 문맥 정보에 대한 식별자를 줬기 때문에 관련 문맥을 사용할 때도 동일한 식별자를 줘야 합니다. 아래는 위와 같은 문맥이 있는 상태에서 질문하는 코드를 보여줍니다.

var questions = new[]
{
    "what is my name?",
    "where do I live?",
    "where is my family from?",
    "where have I travelled?",
    "what do I do for work?",
};

foreach (var q in questions)
{
    var response = await kernel.Memory.SearchAsync(MemoryCollectionName, q).FirstOrDefaultAsync();
    Console.WriteLine(q + " " + response?.Metadata.Text);
}

/* 출력 결과
what is my name? My name is Andrea
where do I live? I currently live in Seattle and have been living there since 2005
where is my family from? My family is from New York
where have I travelled? I visited France and Italy five times since 2015
what do I do for work? I currently work as a tourist operator
*/

오호~~~ 점점 더 예제가 재미있어지는군요. ^^

(첨부 파일은 이 글의 예제 코드를 포함합니다.)




참고로, 이때 문맥으로 저장하는 텍스트는 단순히 문자열로 저장되는 것이 아니고, 우리가 지정한 Embedding backend, 즉 위의 예제에서는 "text-embedding-ada-002" 모델로 벡터값으로 인코딩된 유형으로 저장됩니다. 관련해서는 아래의 문서를 참조하라고 하는군요. ^^

What are Embeddings?
; https://learn.microsoft.com/en-us/semantic-kernel/concepts-ai/embeddings

그러고 보면, 지난 예제에서 "text-embedding-ada-002" 모델을 사용해 "안녕하세요, 책 좀 추천해 주세요." 문자열을 전송했더니 다음과 같은 결과가 나왔는데요,

( 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0, 0 0 0 0 0 1 0 0, 0 0 0, 0 0 0, 0 0, 0 0, 0 0 0, 0 0 0 0,0 0 0, 0 0 0,0,0 0 0 0, 0 0 0 0, 0 0 0,0,0 0, 0,0,0,,0,0,0,0 1 0, 0,0,0,0,0,0, 0,0,0,0,0, 0,0,0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0,0,0, 0,0,0, 0,0,0,0,0,, 0,0,0, 0,0,0,0,0,,0,0,0,0, 1 0,0,0,0,0,0,0,0, 0,0,0,0, 0,0,0, 0,0,0, 0 1 0,0,0,0,0,,, 0,0,0,0, 0,0,0, 0,0, 0,0,0, 0,0, 0,0, 0,0, 0,0,0, 0,0, 0,0,1, 0,0, 0,0, 0,0,0,0, 0,0,0, 0,0, 1,0,0, 0,0, 0,0,0, 0,0,1,0,0,0,0,1, 0,0,0, 0,0,0,1,1,0, 0,0,0,0, 1,0,0,1,1,0, 1,0,0,0,0,0, 1,0,0,0,0,1,0,1,1, 1,0,0,0,1,1, 1,0,1,1,0, 0,0,0,0,0, 0,0, 1, 0,0,1,0,1,1,1, 0,0,0,1,0,1,1, 0,0,0,0, 0,0, 1,0,1,0, 1,1,1,0, 0,0,0, 1,0,0,0,1,1, 1,0,1,1,1, 0,0,0,0, 0,0,0,1,1,1,1,1,1,1,1, 1,0,0,0,0, 1,0,1,1,1,1, 0,0,0,0, 1,1,0,1,0,1,1, 1,0,0,0, 0,0,0,1,,1,0,0,0,1,1,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,0,1,0,0,0,1,0,0,0,1,1,1,1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,1,0,1,0,0,0,1,1,1,1,1,1,1,1,1,1,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1

아마도 저것이 "vectors or arrays of numbers that represent the meaning and the context of tokens processed by the model."인 듯합니다.




[이 글에 대해서 여러분들과 의견을 공유하고 싶습니다. 틀리거나 미흡한 부분 또는 의문 사항이 있으시면 언제든 댓글 남겨주십시오.]







[최초 등록일: ]
[최종 수정일: 7/8/2024]

Creative Commons License
이 저작물은 크리에이티브 커먼즈 코리아 저작자표시-비영리-변경금지 2.0 대한민국 라이센스에 따라 이용하실 수 있습니다.
by SeongTae Jeong, mailto:techsharer at outlook.com

비밀번호

댓글 작성자
 



2023-06-01 08시59분
정성태

... 76  77  78  79  80  81  82  83  84  85  86  87  88  89  [90]  ...
NoWriterDateCnt.TitleFile(s)
11685정성태9/6/201818505사물인터넷: 40. 이어폰 소리를 capacitor로 필터링파일 다운로드1
11684정성태9/6/201821118개발 환경 구성: 396. pagefile.sys를 비활성화시켰는데도 working set 메모리가 줄어드는 이유파일 다운로드1
11683정성태9/5/201818751개발 환경 구성: 395. Azure Web App의 이벤트 로그를 확인하는 방법
11682정성태9/5/201817718오류 유형: 484. Fakes를 포함한 단위 테스트 프로젝트를 빌드 시 CS1729 관련 오류 발생
11681정성태9/5/201820407Windows: 149. 다른 컴퓨터의 윈도우 이벤트 로그를 구독하는 방법 [2]
11680정성태9/2/201822569Graphics: 21. shader - _Time 내장 변수를 이용한 UV 변동 효과파일 다운로드1
11679정성태8/30/201820588.NET Framework: 792. C# COM 서버가 제공하는 COM 이벤트를 C++에서 받는 방법 [1]파일 다운로드1
11678정성태8/29/201819022오류 유형: 483. 닷넷 - System.InvalidProgramException [1]
11677정성태8/29/201816748오류 유형: 482. TFS - Could not find a part of the path '...\packages\Microsoft.AspNet.WebApi.5.2.5\.signature.p7s'.
11676정성태8/29/201827593.NET Framework: 791. C# - ElasticSearch를 위한 Client 라이브러리 제작 [1]파일 다운로드1
11675정성태8/29/201817772오류 유형: 481. The located assembly's manifest definition does not match the assembly reference.
11674정성태8/29/201819751Phone: 12. Xamarin - 기존 리모컨 기능을 핸드폰의 적외선 송신으로 구현파일 다운로드1
11673정성태8/28/201817026오류 유형: 480. Fritzing 실행 시 Ordinal Not Found 오류
11672정성태8/28/201817456오류 유형: 479. 윈도우 - 시스템 설정에서 도메인 참가를 위한 "Change" 버튼이 비활성화된 경우
11671정성태8/28/201823835사물인터넷: 39. 아두이노에서 적외선 송신기 기본 사용법파일 다운로드1
11670정성태8/28/201822070사물인터넷: 38. 아두이노에서 적외선 수신기 기본 사용법 [1]파일 다운로드1
11669정성태8/24/201820852개발 환경 구성: 394. 윈도우 환경에서 elasticsearch의 한글 블로그 검색 인덱스 구성
11668정성태8/24/201831880오류 유형: 478. 윈도우 업데이트(KB4458842) 이후 SQL Server 서비스 시작 오류
11667정성태8/24/201818653오류 유형: 477. "Use Unicode UTF-8 for worldwide language support" 옵션 설정 시 SQL Server 2016 설치 오류 [1]
11666정성태8/22/201818555사물인터넷: 37. 아두이노 - 코딩으로 대신하는 오실레이터 회로의 소리 출력파일 다운로드1
11665정성태8/22/201821247사물인터넷: 36. 오실레이터 회로 동작을 아두이노의 코딩으로 구현하는 방법파일 다운로드1
11664정성태8/22/201820874개발 환경 구성: 393. 윈도우 환경에서 elasticsearch의 한글 형태소 분석기 설치 [1]
11663정성태8/22/201823604개발 환경 구성: 392. 윈도우 환경에서 curl.exe를 이용한 elasticsearch 6.x 기본 사용법
11662정성태8/21/201817255사물인터넷: 35. 병렬 회로에서의 커패시터파일 다운로드1
11661정성태8/21/201819561사물인터넷: 34. 트랜지스터 동작 - 컬렉터-이미터 간의 저항 측정파일 다운로드1
11660정성태8/19/201818646사물인터넷: 33. 세라믹 커패시터의 동작 방식파일 다운로드1
... 76  77  78  79  80  81  82  83  84  85  86  87  88  89  [90]  ...