Commit
·
37dd96b
1
Parent(s):
7e3b8dc
fix: 줄바꿈
Browse files- README.md +4 -0
- README_en.md +4 -0
README.md
CHANGED
|
@@ -26,18 +26,22 @@ AI 에이전트의 능력이 고도화되면서, 그 성능을 실제 환경과
|
|
| 26 |
|
| 27 |
# Ko-AgentBench 핵심 특징 ✨
|
| 28 |
**1. 단계별 태스크 설계**
|
|
|
|
| 29 |
단순 도구 호출부터 장기적 맥락 능력, 강건성 처리 능력까지 에이전트의 능력을 7단계로 입체적으로 분석하였습니다.
|
| 30 |
|
| 31 |
**2. 18가지 한국형 API 사용 및 실생활 환경에 특화된 고품질 시나리오 구성**
|
|
|
|
| 32 |
네이버, 지도, 카카오, 웹사이트 등 한국 실사용 환경 기반의 API를 기반으로 국내 사용자의 일상과 밀접한 '약속 예약', '블로그 후기 검색'과 같은 현실적인 문제 해결 시나리오를 구현했습니다.
|
| 33 |
|
| 34 |
**3. 캐시 기반 반복 평가 및 강건성 테스트**
|
|
|
|
| 35 |
'정보 속성 불일치성 변경' 등 기존 벤치마크의 고질적 문제를 해결합니다.
|
| 36 |
실패 API 응답을 개선함에 따라 벤치마크의 일관성과 신뢰도를 보장합니다.
|
| 37 |
|
| 38 |
의도된 오류 상황에서의 오류 인식/대응 능력(전략)까지 평가함으로 현실 환경에서도 안정적으로 작동하는 모델을 선별합니다.
|
| 39 |
|
| 40 |
**4. 단계별 고유 정밀 지표**
|
|
|
|
| 41 |
도구 선택, 파라미터 구성, 데이터 흐름 등 문제 해결의 불필요/소요를 단계별로 평가합니다. 이를 통해 모델의 강/약점 정량적으로 식별합니다.
|
| 42 |
|
| 43 |
## **데이터 로드**
|
|
|
|
| 26 |
|
| 27 |
# Ko-AgentBench 핵심 특징 ✨
|
| 28 |
**1. 단계별 태스크 설계**
|
| 29 |
+
|
| 30 |
단순 도구 호출부터 장기적 맥락 능력, 강건성 처리 능력까지 에이전트의 능력을 7단계로 입체적으로 분석하였습니다.
|
| 31 |
|
| 32 |
**2. 18가지 한국형 API 사용 및 실생활 환경에 특화된 고품질 시나리오 구성**
|
| 33 |
+
|
| 34 |
네이버, 지도, 카카오, 웹사이트 등 한국 실사용 환경 기반의 API를 기반으로 국내 사용자의 일상과 밀접한 '약속 예약', '블로그 후기 검색'과 같은 현실적인 문제 해결 시나리오를 구현했습니다.
|
| 35 |
|
| 36 |
**3. 캐시 기반 반복 평가 및 강건성 테스트**
|
| 37 |
+
|
| 38 |
'정보 속성 불일치성 변경' 등 기존 벤치마크의 고질적 문제를 해결합니다.
|
| 39 |
실패 API 응답을 개선함에 따라 벤치마크의 일관성과 신뢰도를 보장합니다.
|
| 40 |
|
| 41 |
의도된 오류 상황에서의 오류 인식/대응 능력(전략)까지 평가함으로 현실 환경에서도 안정적으로 작동하는 모델을 선별합니다.
|
| 42 |
|
| 43 |
**4. 단계별 고유 정밀 지표**
|
| 44 |
+
|
| 45 |
도구 선택, 파라미터 구성, 데이터 흐름 등 문제 해결의 불필요/소요를 단계별로 평가합니다. 이를 통해 모델의 강/약점 정량적으로 식별합니다.
|
| 46 |
|
| 47 |
## **데이터 로드**
|
README_en.md
CHANGED
|
@@ -27,18 +27,22 @@ To address this issue, we have developed a high-quality agent benchmark speciali
|
|
| 27 |
|
| 28 |
# Ko-AgentBench Key Features ✨
|
| 29 |
**1. Step-by-step Task Design**
|
|
|
|
| 30 |
We have comprehensively analyzed agent capabilities across 7 levels, from simple tool calls to long-term contextual abilities and robustness handling capabilities.
|
| 31 |
|
| 32 |
**2. 18 Korean-specific APIs and High-quality Scenarios Tailored to Real-life Environments**
|
|
|
|
| 33 |
Based on APIs from Korean real-world usage environments such as Naver, Maps, Kakao, and websites, we have implemented realistic problem-solving scenarios closely related to domestic users' daily lives, such as 'appointment booking' and 'blog review search'.
|
| 34 |
|
| 35 |
**3. Cache-based Iterative Evaluation and Robustness Testing**
|
|
|
|
| 36 |
We solve chronic problems of existing benchmarks, such as 'information attribute inconsistency changes'.
|
| 37 |
By improving failed API responses, we ensure benchmark consistency and reliability.
|
| 38 |
|
| 39 |
By evaluating error recognition/response capabilities (strategies) in intentional error situations, we select models that operate stably even in real-world environments.
|
| 40 |
|
| 41 |
**4. Step-specific Precision Metrics**
|
|
|
|
| 42 |
We evaluate the necessity/requirements of problem-solving step by step, including tool selection, parameter configuration, and data flow. Through this, we quantitatively identify the strengths and weaknesses of models.
|
| 43 |
|
| 44 |
## **Data Loading**
|
|
|
|
| 27 |
|
| 28 |
# Ko-AgentBench Key Features ✨
|
| 29 |
**1. Step-by-step Task Design**
|
| 30 |
+
|
| 31 |
We have comprehensively analyzed agent capabilities across 7 levels, from simple tool calls to long-term contextual abilities and robustness handling capabilities.
|
| 32 |
|
| 33 |
**2. 18 Korean-specific APIs and High-quality Scenarios Tailored to Real-life Environments**
|
| 34 |
+
|
| 35 |
Based on APIs from Korean real-world usage environments such as Naver, Maps, Kakao, and websites, we have implemented realistic problem-solving scenarios closely related to domestic users' daily lives, such as 'appointment booking' and 'blog review search'.
|
| 36 |
|
| 37 |
**3. Cache-based Iterative Evaluation and Robustness Testing**
|
| 38 |
+
|
| 39 |
We solve chronic problems of existing benchmarks, such as 'information attribute inconsistency changes'.
|
| 40 |
By improving failed API responses, we ensure benchmark consistency and reliability.
|
| 41 |
|
| 42 |
By evaluating error recognition/response capabilities (strategies) in intentional error situations, we select models that operate stably even in real-world environments.
|
| 43 |
|
| 44 |
**4. Step-specific Precision Metrics**
|
| 45 |
+
|
| 46 |
We evaluate the necessity/requirements of problem-solving step by step, including tool selection, parameter configuration, and data flow. Through this, we quantitatively identify the strengths and weaknesses of models.
|
| 47 |
|
| 48 |
## **Data Loading**
|