--- language: - ko license: apache-2.0 task_categories: - question-answering tags: - agent - benchmark - tool-use - korean --- # Ko-AgentBench 한국어 에이전트 벤치마크 데이터셋 ## 데이터 로드 ```python from datasets import load_dataset # 특정 레벨 로드 dataset = load_dataset("Hugging-Face-KREW/Ko-AgentBench", data_files="L1.json") # 또는 모든 레벨 로드 dataset = load_dataset("Hugging-Face-KREW/Ko-AgentBench", data_files="*.json") ``` ## 레벨 - L1: Single Tool - L2: Tool Retrieval & Selection - L3: Multiple Tools - L4: Complex Tool Chaining - L5: Conditional Logic - L6: State Management - L7: Multi-step Planning