

기계 번역으로 제공되는 번역입니다. 제공된 번역과 원본 영어의 내용이 상충하는 경우에는 영어 버전이 우선합니다.

# 지도 미세 조정으로 Amazon Nova 모델 미세 조정
<a name="nova-2-sft-data-prep"></a>

Amazon Nova 2.0 SFT 데이터는 선택적 추론 콘텐츠 필드를 추가하여 Amazon Nova 1.0과 동일한 Converse API 형식을 사용합니다. 전체 형식 사양은 [ReasoningContentBlock](https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ReasoningContentBlock.html) 및 [Converse API 스키마](https://docs.aws.amazon.com/bedrock/latest/userguide/conversation-inference-call.html)를 참조하세요.

**지원되는 기능:**
+ **입력 유형** - 사용자 콘텐츠 블록의 텍스트, 이미지 또는 비디오
+ **어시스턴트 콘텐츠** - 텍스트 전용 응답 및 추론 콘텐츠
+ **데이터세트 구성** - 동종이어야 합니다. 텍스트 전용 회전, 텍스트 \+ 이미지 회전 또는 텍스트 \+ 비디오 회전 중 하나를 선택합니다.

**중요**  
동일한 데이터세트 내에서 또는 다른 턴에 이미지와 비디오를 혼합할 수 없습니다.

**현재 제한 사항:**
+ **도구 사용** - 도구 사용은 입력 형식으로 지원되지만 현재 Amazon Nova 2.0 SFT에서는 지원되지 않습니다. 도구 섹션을 추가하면 작업이 실패할 수 있습니다.
+ **멀티모달 추론 콘텐츠** - Converse 형식은 이미지 기반 추론 콘텐츠를 지원하지만 Amazon Nova 2.0 SFT에서는 지원되지 않습니다.
+ **검증 세트** - 검증 세트 제공은 UI를 통해 지원되지만 SFT 훈련 중에는 지원되지 않습니다.

**지원되는 미디어 형식:**
+ **이미지** - PNG, JPEG, GIF
+ **비디오** - MOV, MKV, MP4

## 데이터 형식 예제
<a name="nova-2-sft-data-examples"></a>

------
#### [ Text-only ]

이 예제는 Amazon Nova 1.0과 호환되는 기본 텍스트 전용 형식을 보여줍니다.

```
{
  "schemaVersion": "bedrock-conversation-2024",
  "system": [
    {
      "text": "You are a digital assistant with a friendly personality"
    }
  ],
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "text": "What country is right next to Australia?"
        }
      ]
    },
    {
      "role": "assistant",
      "content": [
        {
          "text": "The closest country is New Zealand"
        }
      ]
    }
  ]
}
```

------
#### [ Text with reasoning ]

이 예제에서는 Amazon Nova 2.0에 대한 선택적 추론 콘텐츠가 포함된 텍스트를 보여줍니다.

```
{
  "schemaVersion": "bedrock-conversation-2024",
  "system": [
    {
      "text": "You are a digital assistant with a friendly personality"
    }
  ],
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "text": "What country is right next to Australia?"
        }
      ]
    },
    {
      "role": "assistant",
      "content": [
        {
          "reasoningContent": {
            "reasoningText": {
              "text": "I need to use my world knowledge of geography to answer this question"
            }
          }
        },
        {
          "text": "The closest country to Australia is New Zealand, located to the southeast across the Tasman Sea."
        }
      ]
    }
  ]
}
```

**참고**  
현재 `reasoningText`는 내에서만 지원됩니다`reasoningContent`. 다중 모달 추론 콘텐츠는 아직 사용할 수 없습니다.

------
#### [ Image \+ text ]

이 예제에서는 텍스트와 함께 이미지 입력을 포함하는 방법을 보여줍니다.

```
{
  "schemaVersion": "bedrock-conversation-2024",
  "system": [
    {
      "text": "You are a helpful assistant."
    }
  ],
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "image": {
            "format": "jpeg",
            "source": {
              "s3Location": {
                "uri": "s3://your-bucket/your-path/your-image.jpg",
                "bucketOwner": "your-aws-account-id"
              }
            }
          }
        },
        {
          "text": "Which country is highlighted in the image?"
        }
      ]
    },
    {
      "role": "assistant",
      "content": [
        {
          "reasoningContent": {
            "reasoningText": {
              "text": "I will determine the highlighted country by examining its location on the map and using my geographical knowledge"
            }
          }
        },
        {
          "text": "The highlighted country is New Zealand"
        }
      ]
    }
  ]
}
```

------
#### [ Video \+ text ]

이 예제에서는 텍스트와 함께 비디오 입력을 포함하는 방법을 보여줍니다.

```
{
  "schemaVersion": "bedrock-conversation-2024",
  "system": [
    {
      "text": "You are a helpful assistant."
    }
  ],
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "video": {
            "format": "mp4",
            "source": {
              "s3Location": {
                "uri": "s3://your-bucket/your-path/your-video.mp4",
                "bucketOwner": "your-aws-account-id"
              }
            }
          }
        },
        {
          "text": "What is shown in this video?"
        }
      ]
    },
    {
      "role": "assistant",
      "content": [
        {
          "reasoningContent": {
            "reasoningText": {
              "text": "I will analyze the video content to identify key elements"
            }
          }
        },
        {
          "text": "The video shows a map with New Zealand highlighted"
        }
      ]
    }
  ]
}
```

------

## 추론 콘텐츠
<a name="nova-2-reasoning-modes"></a>

추론 콘텐츠(연쇄적 사고라고도 함)는 최종 답변을 생성하기 전에 모델의 중간 사고 단계를 캡처합니다. `assistant` 턴에서 `reasoningContent` 필드를 사용하여 이러한 추론 트레이스를 포함합니다.

**손실 계산 방법:**
+ **추론 콘텐츠 포함** - 훈련 손실에 추론 토큰과 최종 출력 토큰이 모두 포함됨
+ **추론 콘텐츠 제외** - 훈련 손실은 최종 출력 토큰에서만 계산됨

**추론 모드를 활성화하는 경우:** 모델이 최종 출력을 생성하기 전에 사고 토큰을 생성하도록 하거나 복잡한 추론 작업`reasoning_enabled: true`에서 향상된 성능이 필요한 경우 훈련 구성에서를 설정합니다. 명시적 추론 단계의 이점을 얻지 못하는 간단한 작업을 훈련할 `reasoning_enabled: false` 때를 설정합니다.

**참고**  
훈련 데이터에 추론 콘텐츠가 포함되어 있는지 여부에 관계없이 추론 모드를 활성화할 수 있습니다. 그러나 모델이 이러한 예제에서 학습하고 추론 품질을 개선할 수 있도록 훈련 데이터에 추론 추적을 포함하는 것이 좋습니다.

**형식 지정 지침:**
+ 추론 콘텐츠에는 일반 텍스트를 사용합니다.
+ 작업에 특별히 필요하지 `</thinking>` 않은 한 `<thinking>` 및와 같은 마크업 태그를 사용하지 마세요.
+ 추론 콘텐츠가 명확하고 문제 해결 프로세스와 관련이 있는지 확인합니다.

**효과적인 추론 콘텐츠에는 다음이 포함되어야 합니다.**
+ 중간 사고 및 분석
+ 논리적 연역 및 추론 단계
+ 단계별 문제 해결 접근 방식
+ 단계와 결론 사이에서 명시적 연결

데이터세트에 추론 트레이스가 없는 경우 Nova Premier와 같은 추론 지원 모델을 사용하여 생성할 수 있습니다. 모델에 입력 및 출력 페어를 제공하고 해당 추론 프로세스를 캡처하여 추론 증강 데이터세트를 구축합니다.

## 데이터세트 준비 지침
<a name="nova-2-dataset-preparation"></a>

다음 표에는 훈련 데이터 세트를 준비하기 위한 지침이 나와 있습니다.


**데이터세트 준비 지침**  

| 지침 | 설명 | 
| --- | --- | 
| 크기 및 품질 |  [See the AWS documentation website for more details](http://docs.aws.amazon.com/ko_kr/bedrock/latest/userguide/nova-2-sft-data-prep.html)  | 
| 다양성 | 다음을 수행하는 다양한 예제를 포함합니다.[See the AWS documentation website for more details](http://docs.aws.amazon.com/ko_kr/bedrock/latest/userguide/nova-2-sft-data-prep.html) | 
| 출력 형식 | 어시스턴트 응답에서 원하는 출력 형식을 명확하게 지정합니다. 예를 들어 JSON 구조, 테이블, CSV 형식 또는 애플리케이션별 사용자 지정 형식이 있습니다. | 
| 멀티턴 대화 |  [See the AWS documentation website for more details](http://docs.aws.amazon.com/ko_kr/bedrock/latest/userguide/nova-2-sft-data-prep.html)  | 
| 품질 체크리스트 |  [See the AWS documentation website for more details](http://docs.aws.amazon.com/ko_kr/bedrock/latest/userguide/nova-2-sft-data-prep.html)  | 