xiaoluodeng 4n3mone commited on
Commit
2ad0c52
ยท
verified ยท
0 Parent(s):

Duplicate from huggingface-KREW/Ko-AgentBench

Browse files

Co-authored-by: yongsang yoo <4n3mone@users.noreply.huggingface.co>

.gitattributes ADDED
@@ -0,0 +1,59 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.lz4 filter=lfs diff=lfs merge=lfs -text
12
+ *.mds filter=lfs diff=lfs merge=lfs -text
13
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
14
+ *.model filter=lfs diff=lfs merge=lfs -text
15
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
16
+ *.npy filter=lfs diff=lfs merge=lfs -text
17
+ *.npz filter=lfs diff=lfs merge=lfs -text
18
+ *.onnx filter=lfs diff=lfs merge=lfs -text
19
+ *.ot filter=lfs diff=lfs merge=lfs -text
20
+ *.parquet filter=lfs diff=lfs merge=lfs -text
21
+ *.pb filter=lfs diff=lfs merge=lfs -text
22
+ *.pickle filter=lfs diff=lfs merge=lfs -text
23
+ *.pkl filter=lfs diff=lfs merge=lfs -text
24
+ *.pt filter=lfs diff=lfs merge=lfs -text
25
+ *.pth filter=lfs diff=lfs merge=lfs -text
26
+ *.rar filter=lfs diff=lfs merge=lfs -text
27
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
28
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
29
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
30
+ *.tar filter=lfs diff=lfs merge=lfs -text
31
+ *.tflite filter=lfs diff=lfs merge=lfs -text
32
+ *.tgz filter=lfs diff=lfs merge=lfs -text
33
+ *.wasm filter=lfs diff=lfs merge=lfs -text
34
+ *.xz filter=lfs diff=lfs merge=lfs -text
35
+ *.zip filter=lfs diff=lfs merge=lfs -text
36
+ *.zst filter=lfs diff=lfs merge=lfs -text
37
+ *tfevents* filter=lfs diff=lfs merge=lfs -text
38
+ # Audio files - uncompressed
39
+ *.pcm filter=lfs diff=lfs merge=lfs -text
40
+ *.sam filter=lfs diff=lfs merge=lfs -text
41
+ *.raw filter=lfs diff=lfs merge=lfs -text
42
+ # Audio files - compressed
43
+ *.aac filter=lfs diff=lfs merge=lfs -text
44
+ *.flac filter=lfs diff=lfs merge=lfs -text
45
+ *.mp3 filter=lfs diff=lfs merge=lfs -text
46
+ *.ogg filter=lfs diff=lfs merge=lfs -text
47
+ *.wav filter=lfs diff=lfs merge=lfs -text
48
+ # Image files - uncompressed
49
+ *.bmp filter=lfs diff=lfs merge=lfs -text
50
+ *.gif filter=lfs diff=lfs merge=lfs -text
51
+ *.png filter=lfs diff=lfs merge=lfs -text
52
+ *.tiff filter=lfs diff=lfs merge=lfs -text
53
+ # Image files - compressed
54
+ *.jpg filter=lfs diff=lfs merge=lfs -text
55
+ *.jpeg filter=lfs diff=lfs merge=lfs -text
56
+ *.webp filter=lfs diff=lfs merge=lfs -text
57
+ # Video files - compressed
58
+ *.mp4 filter=lfs diff=lfs merge=lfs -text
59
+ *.webm filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,224 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - ko
4
+ license: apache-2.0
5
+ task_categories:
6
+ - question-answering
7
+ tags:
8
+ - agent
9
+ - benchmark
10
+ - tool-use
11
+ - korean
12
+ configs:
13
+ - config_name: default
14
+ data_files:
15
+ - split: L1
16
+ path: data/L1-*
17
+ - split: L2
18
+ path: data/L2-*
19
+ - split: L3
20
+ path: data/L3-*
21
+ - split: L4
22
+ path: data/L4-*
23
+ - split: L5
24
+ path: data/L5-*
25
+ - split: L6
26
+ path: data/L6-*
27
+ - split: L7
28
+ path: data/L7-*
29
+ dataset_info:
30
+ features:
31
+ - name: instruction
32
+ dtype: string
33
+ - name: tools
34
+ list: string
35
+ splits:
36
+ - name: L1
37
+ num_bytes: 1551
38
+ num_examples: 11
39
+ - name: L2
40
+ num_bytes: 4655
41
+ num_examples: 30
42
+ - name: L3
43
+ num_bytes: 1433
44
+ num_examples: 10
45
+ - name: L4
46
+ num_bytes: 1567
47
+ num_examples: 10
48
+ - name: L5
49
+ num_bytes: 2091
50
+ num_examples: 20
51
+ - name: L6
52
+ num_bytes: 1184
53
+ num_examples: 15
54
+ - name: L7
55
+ num_bytes: 1302
56
+ num_examples: 10
57
+ download_size: 20447
58
+ dataset_size: 13783
59
+ ---
60
+
61
+ <p align="center">
62
+ <img src="banner.png" />
63
+ </p>
64
+
65
+ # **๐Ÿ‡ฐ๐Ÿ‡ท Ko-AgentBench v1**
66
+
67
+ **"ํ•œ๊ตญ ์—์ด์ „ํŠธ ๋ฒค์น˜๋งˆํฌ ํ”„๋กœ์ ํŠธ"**
68
+
69
+ **[English](README_en.md) | ํ•œ๊ตญ์–ด**
70
+
71
+ <div align="center">
72
+
73
+ [![๐Ÿ† Leaderboard](https://img.shields.io/badge/๐Ÿ†-Leaderboard-blue)](https://huggingface.co/spaces/huggingface-KREW/Ko-AgentBench)
74
+ [![๐Ÿ’ป GitHub](https://img.shields.io/badge/๐Ÿ’ป-GitHub-black)](https://github.com/Hugging-Face-KREW/Ko-AgentBench)
75
+ [![๐Ÿ“Š Dataset](https://img.shields.io/badge/๐Ÿ“Š-Dataset-yellow)](https://huggingface.co/datasets/huggingface-KREW/Ko-AgentBench)
76
+
77
+ </div>
78
+
79
+ ---
80
+
81
+ > **โš ๏ธ ๋ฒค์น˜๋งˆํฌ ํ‰๊ฐ€๋ฅผ ์ง„ํ–‰ํ•˜์‹œ๋ ค๋ฉด [GitHub Repository](https://github.com/Hugging-Face-KREW/Ko-AgentBench)๋ฅผ ๋ฐฉ๋ฌธํ•ด์ฃผ์„ธ์š”.**
82
+ >
83
+ > ์ด ๋ฐ์ดํ„ฐ์…‹์€ ๋ฒค์น˜๋งˆํฌ ํƒœ์Šคํฌ ์ •๋ณด๋งŒ ํฌํ•จํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค์ œ ํ‰๊ฐ€ ์ฝ”๋“œ, API ๋„๊ตฌ, ํ‰๊ฐ€ ๋ฉ”ํŠธ๋ฆญ ๋“ฑ์€ GitHub์—์„œ ํ™•์ธํ•˜์‹ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
84
+
85
+ ---
86
+
87
+ AI ์—์ด์ „ํŠธ์˜ ๋Šฅ๋ ฅ์ด ๊ณ ๋„ํ™”๋˜๋ฉด์„œ, ๊ทธ ์„ฑ๋Šฅ์„ ์‹ค์ œ ํ™˜๊ฒฝ๊ณผ ์œ ์‚ฌํ•œ ์กฐ๊ฑด์—์„œ ์ •๋ฐ€ํ•˜๊ฒŒ ์ธก์ •ํ•˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ด์กŒ์Šต๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ๋Œ€๋ถ€๋ถ„์˜ ๋ฒค์น˜๋งˆํฌ๋Š” ์˜์–ด๊ถŒ ํ™˜๊ฒฝ์„ ๊ธฐ์ค€์œผ๋กœ ์„ค๊ณ„๋˜์–ด, ํ•œ๊ตญ์˜ ํŠน์ˆ˜ํ•œ ์‚ฌ์šฉ ๋งฅ๋ฝ์„ ๋ฐ˜์˜ํ•˜๋Š” ๋ฐ ํ•œ๊ณ„๊ฐ€ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค.
88
+ ์ด๋Ÿฌํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•ด, ํ•œ๊ตญ ์‹ค์‚ฌ์šฉ ํ™˜๊ฒฝ์— ํŠนํ™”๋œ ๊ณ ํ’ˆ์งˆ ์—์ด์ „ํŠธ ๋ฒค์น˜๋งˆํฌ๋ฅผ ๊ฐœ๋ฐœํ•˜์˜€์Šต๋‹ˆ๋‹ค.
89
+
90
+ # Ko-AgentBench ํ•ต์‹ฌ ํŠน์ง• โœจ
91
+ **1. ๋‹จ๊ณ„๋ณ„ ํƒœ์Šคํฌ ์„ค๊ณ„**
92
+
93
+ ๋‹จ์ˆœ ๋„๊ตฌ ํ˜ธ์ถœ๋ถ€ํ„ฐ ์žฅ๊ธฐ์  ๋งฅ๋ฝ ๋Šฅ๋ ฅ, ๊ฐ•๊ฑด์„ฑ ์ฒ˜๋ฆฌ ๋Šฅ๋ ฅ๊นŒ์ง€ ์—์ด์ „ํŠธ์˜ ๋Šฅ๋ ฅ์„ 7๋‹จ๊ณ„๋กœ ์ž…์ฒด์ ์œผ๋กœ ๋ถ„์„ํ•˜์˜€์Šต๋‹ˆ๋‹ค.
94
+
95
+ **2. 18๊ฐ€์ง€ ํ•œ๊ตญํ˜• API ์‚ฌ์šฉ ๋ฐ ์‹ค์ƒํ™œ ํ™˜๊ฒฝ์— ํŠนํ™”๋œ ๊ณ ํ’ˆ์งˆ ์‹œ๋‚˜๋ฆฌ์˜ค ๊ตฌ์„ฑ**
96
+
97
+ ๋„ค์ด๋ฒ„, ์ง€๋„, ์นด์นด์˜ค, ์›น์‚ฌ์ดํŠธ ๋“ฑ ํ•œ๊ตญ ์‹ค์‚ฌ์šฉ ํ™˜๊ฒฝ ๊ธฐ๋ฐ˜์˜ API๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ตญ๋‚ด ์‚ฌ์šฉ์ž์˜ ์ผ์ƒ๊ณผ ๋ฐ€์ ‘ํ•œ '์•ฝ์† ์˜ˆ์•ฝ', '๋ธ”๋กœ๊ทธ ํ›„๊ธฐ ๊ฒ€์ƒ‰'๊ณผ ๊ฐ™์€ ํ˜„์‹ค์ ์ธ ๋ฌธ์ œ ํ•ด๊ฒฐ ์‹œ๋‚˜๋ฆฌ์˜ค๋ฅผ ๊ตฌํ˜„ํ–ˆ์Šต๋‹ˆ๋‹ค.
98
+
99
+ **3. ์บ์‹œ ๊ธฐ๋ฐ˜ ๋ฐ˜๋ณต ํ‰๊ฐ€ ๋ฐ ๊ฐ•๊ฑด์„ฑ ํ…Œ์ŠคํŠธ**
100
+
101
+ '์ •๋ณด ์†์„ฑ ๋ถˆ์ผ์น˜์„ฑ ๋ณ€๊ฒฝ' ๋“ฑ ๊ธฐ์กด ๋ฒค์น˜๋งˆํฌ์˜ ๊ณ ์งˆ์  ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•ฉ๋‹ˆ๋‹ค.
102
+ ์‹คํŒจ API ์‘๋‹ต์„ ๊ฐœ์„ ํ•จ์— ๋”ฐ๋ผ ๋ฒค์น˜๋งˆํฌ์˜ ์ผ๊ด€์„ฑ๊ณผ ์‹ ๋ขฐ๋„๋ฅผ ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค.
103
+
104
+ ์˜๋„๋œ ์˜ค๋ฅ˜ ์ƒํ™ฉ์—์„œ์˜ ์˜ค๋ฅ˜ ์ธ์‹/๋Œ€์‘ ๋Šฅ๋ ฅ(์ „๋žต)๊นŒ์ง€ ํ‰๊ฐ€ํ•จ์œผ๋กœ ํ˜„์‹ค ํ™˜๊ฒฝ์—์„œ๋„ ์•ˆ์ •์ ์œผ๋กœ ์ž‘๋™ํ•˜๋Š” ๋ชจ๋ธ์„ ์„ ๋ณ„ํ•ฉ๋‹ˆ๋‹ค.
105
+
106
+ **4. ๋‹จ๊ณ„๋ณ„ ๊ณ ์œ  ์ •๋ฐ€ ์ง€ํ‘œ**
107
+
108
+ ๋„๊ตฌ ์„ ํƒ, ํŒŒ๋ผ๋ฏธํ„ฐ ๊ตฌ์„ฑ, ๋ฐ์ดํ„ฐ ํ๋ฆ„ ๋“ฑ ๋ฌธ์ œ ํ•ด๊ฒฐ์˜ ๋ถˆํ•„์š”/์†Œ์š”๋ฅผ ๋‹จ๊ณ„๋ณ„๋กœ ํ‰๊ฐ€ํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ๋ชจ๋ธ์˜ ๊ฐ•/์•ฝ์  ์ •๋Ÿ‰์ ์œผ๋กœ ์‹๋ณ„ํ•ฉ๋‹ˆ๋‹ค.
109
+
110
+ ## **๋ฐ์ดํ„ฐ ๋กœ๋“œ**
111
+
112
+ ```python
113
+ from datasets import load_dataset
114
+
115
+ # ์ „์ฒด ๋ฐ์ดํ„ฐ์…‹ ๋กœ๋“œ
116
+ dataset = load_dataset("huggingface-KREW/Ko-AgentBench")
117
+
118
+ # ํŠน์ • ๋ ˆ๋ฒจ๋งŒ ๋กœ๋“œ
119
+ l1_dataset = load_dataset("huggingface-KREW/Ko-AgentBench", split="L1")
120
+
121
+ # ๋ฐ์ดํ„ฐ ํ™•์ธ
122
+ print(dataset["L1"][0])
123
+ # {
124
+ # 'instruction': 'ํŒ๊ต์—ญ์—์„œ ์ž ์‹ค์•ผ๊ตฌ์žฅ๊นŒ์ง€ ์ž์ฐจ๋กœ ๋ช‡ ๋ถ„ ๊ฑธ๋ฆด๊นŒ?',
125
+ # 'tools': ['Directions_naver']
126
+ # }
127
+ ```
128
+
129
+ # ๋ฐ์ดํ„ฐ์…‹ ๊ฐœ์š”
130
+
131
+ - ์—์ด์ „ํŠธ ๋ฒค์น˜๋งˆํฌ ์„ค๊ณ„๋ฅผ ์œ„ํ•œ ํƒœ์Šคํฌ ๋ถ„๋ฅ˜ ์ฒด๊ณ„ ์ •์˜
132
+ - ์—์ด์ „ํŠธ์˜ Tool calling ํ™œ์šฉํ•˜๋Š” ๊ณผ์ •์—์„œ ํ•„์š”ํ•œ ๋Šฅ๋ ฅ์„ ๋‹จ๊ณ„์ ์œผ๋กœ ๋ถ„๋ฆฌํ•˜์—ฌ ํ‰๊ฐ€ํ•  ์ˆ˜ ์žˆ๋„๋ก ์„ค๊ณ„
133
+
134
+ ## ๋ฐ์ดํ„ฐ์…‹ ๋ฒ”์œ„
135
+
136
+ - ํ‰๊ฐ€ ๋Œ€์ƒ : Open-weight sLLM(*supports tool calling), Commercial APIs
137
+ - ํ‰๊ฐ€ ๋ฒ”์œ„ : ํ‰๊ฐ€ ์˜์—ญ : ๋‹จ์ผํ„ด(single-turn) ๋ฐ ๋ฉ€ํ‹ฐํ„ด(multi-turn) ๋Œ€ํ™” ์ƒํ™ฉ์—์„œ Agent๋กœ์จ Tool calling ์ˆ˜ํ–‰ ๋Šฅ๋ ฅ
138
+ - ์ ์šฉ API : 18๊ฐ€์ง€ ํ•œ๊ตญํ˜• ์˜คํ”ˆAPI
139
+
140
+ # ํƒœ์Šคํฌ ๋ถ„๋ฅ˜ ๋‹จ๊ณ„
141
+
142
+ ## ์‹ฑ๊ธ€ํ„ด
143
+
144
+ **L1. (๋‹จ์ผ ๋„๊ตฌ ํ˜ธ์ถœ) Single Tool Call**
145
+ - ๋ชฉํ‘œ: ๊ฐ€์žฅ ๊ธฐ๋ณธ์ ์ธ API ํ˜ธ์ถœ ๋Šฅ๋ ฅ ๊ฒ€์ฆ
146
+ - ์„ค๋ช…: ์ฃผ์–ด์ง„ ๋„๊ตฌ๋ฅผ ์ •ํ™•ํ•œ ํŒŒ๋ผ๋ฏธํ„ฐ๋กœ ์‹คํ–‰ํ•  ์ˆ˜ ์žˆ๋Š”์ง€ ํ™•์ธ
147
+ - ํŠน์ง•: API๋ช…์ด ๋ช…์‹œ๋œ ์š”์ฒญor ์ž์—ฐ์–ด ์š”์ฒญ์„ ๊ทธ๋Œ€๋กœ ์ˆ˜ํ–‰ โ†’ "์ •ํ™•์„ฑ๋งŒ" ํ‰๊ฐ€
148
+ - ์˜ˆ์‹œ: "๋„ค์ด๋ฒ„ ์ฑ… API๋กœ '๊ธ‰๋ฅ˜'๋ฅผ ๊ฒ€์ƒ‰ํ•˜๏ฟฝ๏ฟฝ๏ฟฝ ๊ฐ€๊ฒฉ ์•Œ๋ ค ์ค˜."
149
+ - ์˜ˆ์‹œ: "๊ธ‰๋ฅ˜ ์ฑ… ๊ฐ€๊ฒฉ ์•Œ๋ ค์ค˜"
150
+
151
+ **L2. (๋„๊ตฌ ์„ ํƒ) Tool Selection**
152
+ - ๋ชฉํ‘œ: ์—ฌ๋Ÿฌ ํ›„๋ณด ๋„๊ตฌ ์ค‘ ์ตœ์ ์˜ API๋ฅผ ์„ ํƒํ•˜๋Š” ๋Šฅ๋ ฅ ๊ฒ€์ฆ
153
+ - ์„ค๋ช…: ์‚ฌ์šฉ์ž๋Š” ์ž์—ฐ์–ด๋กœ ์š”์ฒญํ•˜๊ณ , ๋ชจ๋ธ์€ ์ฃผ์–ด์ง„ ๋„๊ตฌ ๋ชฉ๋ก ์ค‘ ๊ฐ€์žฅ ์ ํ•ฉํ•œ ๋„๊ตฌ๋ฅผ ์„ ํƒํ•ด์•ผ ํ•จ
154
+ - ํŠน์ง•: ์ž…๋ ฅ๋œ ์ž์—ฐ์–ด๋กœ ์ •ํ™•ํ•œ tool mapping ํ‰๊ฐ€
155
+ - ์˜ˆ์‹œ: "'์˜ฌ๋ฐฑ์˜์–ด ์ค‘2-1 ์ฒœ์žฌ(๊น€)' ์ฑ… ๊ฐ€๊ฒฉ ํ™•์ธํ•ด์ค˜."
156
+ - ํ›„๋ณด ๋„๊ตฌ: `hotel_booking_api`, `aladin_books_api`
157
+ - ํ›„๋ณด ๋„๊ตฌ๋Š” ์ƒํ˜ธ ์—ฐ๊ด€์„ฑ์ด ์—†์–ด์•ผ ํ•จ์„ ์กฐ๊ฑด์œผ๋กœ ํ•ฉ๋‹ˆ๋‹ค.
158
+
159
+ **L3 (๋„๊ตฌ ์ˆœ์ฐจ ์ถ”๋ก ) Sequential Tool Reasoning**
160
+ - ๋ชฉํ‘œ: ๋‹ค๋‹จ๊ณ„ reasoning์„ ํ†ตํ•œ ๊ณ„ํš ๋ฐ ์‹คํ–‰ ๋Šฅ๋ ฅ ๊ฒ€์ฆ
161
+ - ์„ค๋ช…: ํ•œ ๋„๊ตฌ์˜ ๊ฒฐ๊ณผ๋ฅผ ๋‹ค๋ฅธ ๋„๊ตฌ ์ž…๋ ฅ์œผ๋กœ ์—ฐ๊ฒฐํ•˜์—ฌ ์˜ฌ๋ฐ”๋ฅธ pipeline์„ ๊ตฌ์„ฑํ•  ์ˆ˜ ์žˆ๋Š”์ง€ ํ™•์ธ
162
+ - ํŠน์ง•: ๋‹จ์ˆœ ํ˜ธ์ถœ์ด ์•„๋‹ˆ๋ผ "๊ณ„ํš์„ฑ ์žˆ๋Š” chain-of-tools" ํ‰๊ฐ€
163
+ - ์˜ˆ์‹œ: "11๋ฒˆ๊ฐ€ ์•„๋งˆ์กด์—์„œ ๊ตฌ๋งคํ•œ ์ธ์Šคํƒ์Šค11 ์–ธ์ œ ๋ฐฐ์†ก์˜ค๋Š”์ง€ ์•Œ๋ ค์ค˜"
164
+ - ํ›„๋ณด ๋„๊ตฌ: `11st_order_api`, `๊ด€์„ธ์ฒญ_api`, `cj_delivery_api`
165
+ - ์ˆœ์ฐจ์ ์œผ๋กœ ๋„๊ตฌ๋ฅผ ํ˜ธ์ถœ ํ•  ์ˆ˜ ์žˆ์–ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.(11๋ฒˆ๊ฐ€์—์„œ ๋ฐฐ์†ก๋ฒˆํ˜ธ ์กฐํšŒโ†’๊ด€์„ธ์ฒญ ํ†ต๊ด€โ†’ํƒ๋ฐฐ์‚ฌ)
166
+
167
+ **L4 (๋„๊ตฌ ๋ณ‘๋ ฌ ์ถ”๋ก ) Parallel Tool Reasoning**
168
+ - ๋ชฉํ‘œ: ๋ณ‘๋ ฌ์ ์œผ๋กœ ์ •๋ณด๋ฅผ ์ˆ˜์ง‘ํ•˜๊ณ , ์ด๋ฅผ ์ข…ํ•ฉํ•˜์—ฌ ๊ฒฐ๋ก  ๋„์ถœ
169
+ - ์„ค๋ช…: ์„œ๋กœ ๋…๋ฆฝ์ ์ธ ์—ฌ๋Ÿฌ ๋„๊ตฌ๋ฅผ ๋™์‹œ์— ํ˜ธ์ถœํ•˜๊ณ , ๊ฒฐ๊ณผ๋ฅผ ๋น„๊ตยท๋ถ„์„ ํ›„ ์ตœ์ข… ๋‹ต๋ณ€ ์‚ฐ์ถœ
170
+ - ํŠน์ง•: Multi-source aggregation ํ‰๊ฐ€ (์ •๋ณด ์ข…ํ•ฉยท๋น„๊ต ๋Šฅ๋ ฅ)
171
+ - ์˜ˆ์‹œ: "'ํ•œ๋กœ๋กœ ์ž๋ชฝ์‚ด๊ตฌํด๋Ÿฝ' ์ฑ… ์žฌ๊ณ  ํ™•์ธํ•ด์ค˜."
172
+ - ํ›„๋ณด ๋„๊ตฌ: `kyobo_books_api`, `aladin_books_api`
173
+ - ์˜ˆ์ƒ ๋‹ต๋ณ€: ๊ต๋ณด๋ฌธ๊ณ ์— 12๊ถŒ, ์•Œ๋ผ๋”˜์— 18๊ถŒ ์ด 30๊ถŒ ์žˆ์Šต๋‹ˆ๋‹ค.
174
+ - ์ด๋•Œ ํ›„๋ณด ๋„๊ตฌ๋Š” ๋ณ‘๋ ฌ์ ์œผ๋กœ ๊ฐ™์€ ๊ธฐ๋Šฅ์„ ๋‹ด๋‹นํ•ด์•ผ ํ•จ.
175
+
176
+ **L5 (์˜ค๋ฅ˜ ์ฒ˜๋ฆฌ์™€ ๊ฐ•๊ฑด์„ฑ) Error Handling and Robustness**
177
+ - ๋ชฉํ‘œ: ์˜ค๋ฅ˜ ์ƒํ™ฉ์—์„œ์˜ ๋Œ€์ฒ˜ ๋Šฅ๋ ฅ ๊ฒ€์ฆ
178
+ - ์„ค๋ช…: ๋‹จ์ˆœํžˆ "์‹คํŒจํ–ˆ๋‹ค"๊ฐ€ ์•„๋‹ˆ๋ผ, ๋‹ค์–‘ํ•œ failure mode๋ฅผ ์–ด๋–ป๊ฒŒ ์ฒ˜๋ฆฌํ•˜๋Š”์ง€ ํ‰๊ฐ€
179
+ - **์„ธ๋ถ€ ํ•ญ๋ชฉ:**
180
+ - A. ์ถ”๊ฐ€ ์งˆ๋ฌธ ์š”์ฒญ
181
+ - ์ •๋ณด ๋ถ€์กฑ ์‹œ ์‚ฌ์šฉ์ž๊ฐ€ ๋” ๋ช…ํ™•ํ•œ ์š”์ฒญ์„ ํ•˜๋„๋ก ์œ ๋„
182
+ - B. Hallucination ๋ฐฉ์ง€
183
+ - ์กด์žฌํ•˜์ง€ ์•Š๋Š” API ํ˜ธ์ถœ ๊ธˆ์ง€
184
+ - ์‹คํŒจ ์‹œ "์„ฑ๊ณตํ•œ ๊ฒƒ์ฒ˜๋Ÿผ ๊พธ๋ฉฐ๋‚ด๋Š” ๋‹ต๋ณ€" ๊ธˆ์ง€
185
+ - C. ํšŒํ”ผ๊ธฐ๋™(Fallback)
186
+ - ํŠน์ • API ์˜ค๋ฅ˜ ์‹œ, ๋™์ผ ๊ธฐ๋Šฅ์„ ๊ฐ€์ง„ ๋Œ€์ฒด API ํ™œ์šฉ ๊ฐ€๋Šฅ ์—ฌ๋ถ€
187
+ - ์˜ˆ์‹œ: "๋„ค์ด๋ฒ„ ์˜ํ™” API ํ˜ธ์ถœ ์‹คํŒจ ์‹œ โ†’ 'API ํ˜ธ์ถœ ์‹คํŒจ' ๋ณด๊ณ  or ์นด์นด์˜ค ์˜ํ™” API ๋Œ€์ฒด ํ˜ธ์ถœ"
188
+
189
+ ## ๋ฉ€ํ‹ฐํ„ด
190
+
191
+ **L6 (ํšจ์œจ์ ์ธ ๋„๊ตฌ ํ™œ์šฉ) Efficient Tool Utilization**
192
+ - ๋ชฉํ‘œ: ์ด์ „ Tool ๊ฒฐ๊ณผ๋ฅผ ํšจ์œจ์ ์œผ๋กœ ์žฌํ™œ์šฉํ•˜๋Š” ๋Šฅ๋ ฅ ๊ฒ€์ฆ
193
+ - ์„ค๋ช…: ๋ชจ๋“  ์ƒํ™ฉ์—์„œ API๋ฅผ ์žฌํ˜ธ์ถœํ•˜๋Š” ๊ฒƒ์€ ์ •ํ™•ํ•˜๋”๋ผ๋„ ๋น„์šฉยท์ง€์—ฐ ์ธก๋ฉด์—์„œ ๋น„ํšจ์œจ์ ์ž„. ๋ฐ˜๋Œ€๋กœ ์˜ค๋ž˜๋œ ์ •๋ณด๋ฅผ ๋ฌด์กฐ๊ฑด ์žฌ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ๋„ ์ •ํ™•์„ฑ์— ๋ฌธ์ œ ๋ฐœ์ƒ.
194
+ - ํŠน์ง•: "์žฌํ˜ธ์ถœ vs ์žฌ์‚ฌ์šฉ" ์‚ฌ์ด์—์„œ ํ•ฉ๋ฆฌ์  ์„ ํƒ์„ ํ•  ์ˆ˜ ์žˆ๋Š”๊ฐ€ ํ‰๊ฐ€
195
+ - ์˜ˆ์‹œ:
196
+ - ์‚ฌ์šฉ์ž: "์ฟ ํŒก๊ณผ ๋„ค์ด๋ฒ„ ๊ฐ€๊ฒฉ ๋น„๊ตํ•ด์ค˜." โ†’ ๊ฒฐ๊ณผ: ์ฟ ํŒก 80, ๋„ค์ด๋ฒ„ 85
197
+ - ์‚ฌ์šฉ์ž: "๋„ค์ด๋ฒ„ ๊ฐ€๊ฒฉ ์–ผ๋งˆ์˜€์ง€?"
198
+ - ์˜ฌ๋ฐ”๋ฅธ ๋‹ต๋ณ€: 85 (๊ณผ๊ฑฐ ์ •๋ณด ํ™œ์šฉ, ๋ถˆํ•„์š”ํ•œ ์žฌํ˜ธ์ถœ ํšŒํ”ผ)
199
+ - ์ž˜๋ชป๋œ ๋‹ต๋ณ€: ๋‹ค์‹œ API ํ˜ธ์ถœ or "๋ชฐ๋ผ์š”"
200
+
201
+ **L7 (์žฅ๊ธฐ ์ปจํ…์ŠคํŠธ ๊ธฐ์–ต) Long-Context Reasoning**
202
+ - ๋ชฉํ‘œ: ๋ฉ€ํ‹ฐํ„ด ๋Œ€ํ™”์—์„œ ์žฅ๊ธฐ์  ๋งฅ๋ฝ์„ ์œ ์ง€ํ•˜๋Š” ๋Šฅ๋ ฅ ๊ฒ€์ฆ
203
+ - ์„ค๋ช…: ๋ช‡ ํ„ด ์ „์˜ ์ •๋ณด๋ฅผ ๊ธฐ์–ตํ•˜๊ณ , ์ƒˆ๋กœ์šด ์งˆ๋ฌธ๊ณผ ์—ฐ๊ฒฐํ•˜์—ฌ ์˜ฌ๋ฐ”๋ฅด๊ฒŒ Tool calling ์ˆ˜ํ–‰
204
+ - ์˜ˆ์‹œ:
205
+ - ์‚ฌ์šฉ์ž ์ฒซ ์งˆ๋ฌธ: "์ œ์ฃผ๋„ ์—ฌํ–‰ ๊ฐˆ ๊ฑฐ์•ผ."
206
+ - ์ดํ›„: "๋‚ ์”จ ์–ด๋•Œ?" โ†’ ์ œ์ฃผ๋„ ๋งฅ๋ฝ์„ ํ™œ์šฉํ•ด ๋‚ ์”จ API ํ˜ธ์ถœ
207
+ - (์ถ”๊ฐ€ ํ„ด) "๋น„ ์˜ค๋ฉด ์šฐ์‚ฐ ์‚ด ์ˆ˜ ์žˆ๋Š” ๊ณณ๋„ ์ฐพ์•„์ค˜." โ†’ ์•ž์„  ์ œ์ฃผ๋„+๋‚ ์”จ ์ปจํ…์ŠคํŠธ ๋ชจ๋‘ ํ™œ์šฉ
208
+
209
+
210
+
211
+ ## ๋งํฌ ๐Ÿ”—
212
+ Ko-AgentBench์— ๋Œ€ํ•œ ๋” ์ž์„ธํ•œ ๋‚ด์šฉ์„ ํ™•์ธ ํ•˜์‹ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
213
+ - ๐Ÿ† [Live Leaderboard](https://huggingface.co/spaces/huggingface-KREW/Ko-AgentBench)
214
+ - ๐Ÿ“Š [Dataset](https://huggingface.co/datasets/huggingface-KREW/Ko-AgentBench)
215
+ - ๐Ÿ“ [Github](https://github.com/Hugging-Face-KREW/Ko-AgentBench)
216
+
217
+
218
+ ## ๋ฌธ์˜ ๐Ÿ“ง
219
+ ๋ฐ์ดํ„ฐ์…‹ ๋ฐ ๋ฒค์น˜๋งˆํฌ์— ๋Œ€ํ•œ ๋ฌธ์˜๊ฐ€ ์žˆ์œผ์‹œ๋‹ค๋ฉด ์—ฐ๋ฝ ์ฃผ์„ธ์š”!
220
+
221
+ Hugging Face KREW๋Š” Hugging Face๋ฅผ ํ†ตํ•ด ์ธ๊ณต์ง€๋Šฅ์„ ๊นŠ์ด ์ดํ•ดํ•˜๊ณ , ์˜คํ”ˆ ์†Œ์Šค์— ๊ธฐ์—ฌํ•˜๊ธฐ ์œ„ํ•ด ๋…ธ๋ ฅํ•˜๋Š” ํ•œ๊ตญ ๋น„์˜๋ฆฌ ๋ฆฌ์„œ์น˜ ์กฐ์ง์ž…๋‹ˆ๋‹ค.
222
+ - โœ๐Ÿป Blog: [KREW-blog](https://hugging-face-krew.github.io/)
223
+ - ๐Ÿฆ HuggingFace Community: [@huggingface-KREW](https://huggingface.co/huggingface-KREW)
224
+ - ๐Ÿ’ผ LinkedIn: [Hugging Face KREW](https://www.linkedin.com/company/hugging-face-krew/)
README_en.md ADDED
@@ -0,0 +1,153 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ license: apache-2.0
5
+ task_categories:
6
+ - question-answering
7
+ tags:
8
+ - agent
9
+ - benchmark
10
+ - tool-use
11
+ - korean
12
+ ---
13
+
14
+ <p align="center">
15
+ <img src="banner.png" />
16
+ </p>
17
+
18
+ # **๐Ÿ‡ฐ๐Ÿ‡ท Ko-AgentBench v1**
19
+
20
+ **"Korean Agent Benchmark Project"**
21
+
22
+ **English | [ํ•œ๊ตญ์–ด](README.md)**
23
+
24
+ As AI agents become more sophisticated, it has become crucial to precisely measure their performance under conditions similar to real-world environments. However, most benchmarks are designed based on English-speaking environments, which limits their ability to reflect Korea's unique usage contexts.
25
+
26
+ To address this issue, we have developed a high-quality agent benchmark specialized for the Korean real-world usage environment.
27
+
28
+ # Ko-AgentBench Key Features โœจ
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**
49
+
50
+ ```bash
51
+ from datasets import load_dataset
52
+
53
+ # Load specific level
54
+ dataset = load_dataset("Hugging-Face-KREW/Ko-AgentBench", data_files="L1.json")
55
+
56
+ # Or load all levels
57
+ dataset = load_dataset("Hugging-Face-KREW/Ko-AgentBench", data_files="*.json")
58
+ ```
59
+
60
+ # Dataset Overview
61
+
62
+ - Define task classification system for agent benchmark design
63
+ - Design to evaluate agent's tool calling capabilities in a step-by-step manner
64
+
65
+ ## Dataset Scope
66
+
67
+ - Evaluation Target: Open-weight sLLM (supports tool calling), Commercial APIs
68
+ - Evaluation Scope: Agent tool calling performance in single-turn and multi-turn conversation situations
69
+ - Applied APIs: 18 Korean-specific open APIs
70
+
71
+
72
+ # Task Levels
73
+
74
+ ## Single-Turn
75
+
76
+ **L1. Single Tool Call**
77
+ - Goal: Verify the most basic API calling capability
78
+ - Description: Check if the given tool can be executed with correct parameters
79
+ - Feature: Evaluate "accuracy only" by performing requests with specified API names or natural language requests as-is
80
+ - Example: "Search for 'Rapid Current' using Naver Book API and tell me the price."
81
+ - Example: "Tell me the price of the 'Rapid Current' book"
82
+
83
+ **L2. Tool Selection**
84
+ - Goal: Verify the ability to select the optimal API among multiple candidate tools
85
+ - Description: Users make requests in natural language, and the model must select the most suitable tool from the given tool list
86
+ - Feature: Evaluate accurate tool mapping with input natural language
87
+ - Example: "Check the price of the 'All Back English Middle 2-1 Cheonjae (Kim)' book."
88
+ - Candidate tools: `hotel_booking_api`, `aladin_books_api`
89
+ - Candidate tools must have no mutual correlation.
90
+
91
+ **L3. Sequential Tool Reasoning**
92
+ - Goal: Verify planning and execution capabilities through multi-step reasoning
93
+ - Description: Check if a correct pipeline can be constructed by connecting the results of one tool as input to another tool
94
+ - Feature: Evaluate "planned chain-of-tools" rather than simple calls
95
+ - Example: "Tell me when the Instax11 I bought from 11st Amazon will be delivered"
96
+ - Candidate tools: `11st_order_api`, `customs_api`, `cj_delivery_api`
97
+ - Tools must be callable sequentially (11st delivery number inquiry โ†’ customs clearance โ†’ courier company)
98
+
99
+ **L4. Parallel Tool Reasoning**
100
+ - Goal: Collect information in parallel and derive conclusions by synthesizing it
101
+ - Description: Simultaneously call multiple independent tools, compare and analyze results, then produce final answers
102
+ - Feature: Evaluate multi-source aggregation (information synthesis and comparison ability)
103
+ - Example: "Check the stock of the 'Hanroro Grapefruit Apricot Club' book."
104
+ - Candidate tools: `kyobo_books_api`, `aladin_books_api`
105
+ - Expected answer: There are 12 books at Kyobo Book Centre and 18 books at Aladin, totaling 30 books.
106
+ - At this time, candidate tools must handle the same function in parallel.
107
+
108
+ **L5. Error Handling and Robustness**
109
+ - Goal: Verify coping ability in error situations
110
+ - Description: Evaluate how various failure modes are handled, not just "failed"
111
+ - **Sub-items:**
112
+ - A. Request for additional questions
113
+ - Guide users to make clearer requests when information is insufficient
114
+ - B. Hallucination prevention
115
+ - Prohibit calling non-existent APIs
116
+ - Prohibit "pretending to succeed" answers when failed
117
+ - C. Fallback maneuvers
118
+ - Whether alternative APIs with the same function can be utilized when specific API errors occur
119
+ - Example: "When Naver Movie API call fails โ†’ Report 'API call failed' or call Kakao Movie API as alternative"
120
+
121
+ ## Multi-Turn
122
+
123
+ **L6. Efficient Tool Utilization**
124
+ - Goal: Verify the ability to efficiently reuse previous tool results
125
+ - Description: While recalling APIs in all situations is accurate, it's inefficient in terms of cost and delay. Conversely, unconditionally reusing old information also causes accuracy problems.
126
+ - Feature: Evaluate whether reasonable choices can be made between "recall vs reuse"
127
+ - Example:
128
+ - User: "Compare Coupang and Naver prices." โ†’ Result: Coupang 80, Naver 85
129
+ - User: "What was the Naver price?"
130
+ - Correct answer: 85 (utilize past information, avoid unnecessary recalls)
131
+ - Wrong answer: Call API again or "I don't know"
132
+
133
+ **L7. Long-Context Reasoning**
134
+ - Goal: Verify the ability to maintain long-term context in multi-turn conversations
135
+ - Description: Remember information from several turns ago and correctly perform tool calling by connecting it with new questions
136
+ - Example:
137
+ - User's first question: "I'm going to travel to Jeju Island."
138
+ - Later: "How's the weather?" โ†’ Call weather API using Jeju Island context
139
+ - (Additional turn) "If it rains, find places where I can buy an umbrella." โ†’ Utilize all previous Jeju Island + weather context
140
+
141
+ ## Links
142
+ You can check more detailed information about Ko-AgentBench.
143
+ - ๐Ÿ† [Live Leaderboard](https://huggingface.co/spaces/huggingface-KREW/Ko-AgentBench)
144
+ - ๐Ÿ“Š [Dataset](https://huggingface.co/datasets/huggingface-KREW/Ko-AgentBench)
145
+ - ๐Ÿ“ [Github](https://github.com/Hugging-Face-KREW/Ko-AgentBench)
146
+
147
+ ## Contact
148
+ If you have any questions about the dataset and benchmark, please contact us!
149
+
150
+ Hugging Face KREW is a Korean non-profit research organization that strives to deeply understand artificial intelligence through Hugging Face and contribute to open source.
151
+ - โœ๐Ÿป Blog: [KREW-blog](https://hugging-face-krew.github.io/)
152
+ - ๐Ÿฆ HuggingFace Community: [@huggingface-KREW](https://huggingface.co/huggingface-KREW)
153
+ - ๐Ÿ’ผ LinkedIn: [Hugging Face KREW](https://www.linkedin.com/company/hugging-face-krew/)
banner.png ADDED

Git LFS Details

  • SHA256: c5803c1fcde54de05ea0e6f0b4e6763399368a0fe1d7842351e6b8058c8d1b8b
  • Pointer size: 132 Bytes
  • Size of remote file: 1.08 MB
data/L1-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:22301f215ccdc3d83ae6ecc656da64ed9647739509bcdcf1e67898d3b35858c6
3
+ size 2979
data/L2-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c41f6cf07a2119d0430a30477361cd4fae102dad303bed09f2637bf955d472af
3
+ size 3385
data/L3-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b311828c9e5cc62d4fdeaf3232e93d3ca0dec41cdc469ba70f1ecf0690304ca9
3
+ size 2779
data/L4-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f12956e763a0033955ebd70622ec5533937f2d223def77acc1c200e37982a12c
3
+ size 2817
data/L5-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:df807b23120c5c5d97dbf943cbd4f42f5c7b4c875319d8bc2e81a4cdb6c740bc
3
+ size 3250
data/L6-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:40156957e18dfa1940c8299fa1a91b35b82eb9afbbbe18ffb4f4452723d9fd5c
3
+ size 2433
data/L7-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:40dcf1a7e91ce2dcc37c1663a5eaec28a6b2e0ae50fa02a3f2f0cb99b74e7eb2
3
+ size 2804