Hugging Face's logo

KDLAI
/
KDL-Frontier-Parser-nano-v2

Image-Text-to-Text
Safetensors
Korean
English
qwen2_vl
document-parsing
ocr
table-recognition
chart-extraction
vlm
parsebench
conversational
Eval Results
Model card Files
xet
 Community
1

Add official ParseBench results for KDL-Frontier-Parser-nano-v2

#1
by Haigini - opened
base: refs/heads/main
←
from: refs/pr/1
Discussion Files changed
+72
-0
Files changed (1) hide show
  1. .eval_results/parsebench.yaml +72 -0
.eval_results/parsebench.yaml ADDED
@@ -0,0 +1,72 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ - dataset:
2
+ id: llamaindex/ParseBench
3
+ task_id: mean
4
+ revision: 2805a1d940f95a203e0ae4b88be9934f7765b3fc
5
+ value: 78.95935214038234
6
+ date: '2026-10-28'
7
+ source:
8
+ url: https://huggingface.co/KDLAI/KDL-Frontier-Parser-nano-v2/blob/5b5e0edda3c019563a2f6205ec49095288fafeb8/evaluation/parsebench-20261028-gpu5.json
9
+ name: Official ParseBench full-set author validation
10
+ user: Haigini
11
+ notes: Official full-set single pass; author-measured. Model 766505d578be5d99f2358ec9c95224023cd28d5f; harness 7de04fc11c6a692dfc5e1a86c7be817f3f0af8de.
12
+ See source for coverage.
13
+ - dataset:
14
+ id: llamaindex/ParseBench
15
+ task_id: table
16
+ revision: 2805a1d940f95a203e0ae4b88be9934f7765b3fc
17
+ value: 86.06003056212998
18
+ date: '2026-10-28'
19
+ source:
20
+ url: https://huggingface.co/KDLAI/KDL-Frontier-Parser-nano-v2/blob/5b5e0edda3c019563a2f6205ec49095288fafeb8/evaluation/parsebench-20261028-gpu5.json
21
+ name: Official ParseBench full-set author validation
22
+ user: Haigini
23
+ notes: Official full-set single pass; author-measured. Model 766505d578be5d99f2358ec9c95224023cd28d5f; harness 7de04fc11c6a692dfc5e1a86c7be817f3f0af8de.
24
+ See source for coverage.
25
+ - dataset:
26
+ id: llamaindex/ParseBench
27
+ task_id: chart
28
+ revision: 2805a1d940f95a203e0ae4b88be9934f7765b3fc
29
+ value: 68.87017813954434
30
+ date: '2026-10-28'
31
+ source:
32
+ url: https://huggingface.co/KDLAI/KDL-Frontier-Parser-nano-v2/blob/5b5e0edda3c019563a2f6205ec49095288fafeb8/evaluation/parsebench-20261028-gpu5.json
33
+ name: Official ParseBench full-set author validation
34
+ user: Haigini
35
+ notes: Official full-set single pass; author-measured. Model 766505d578be5d99f2358ec9c95224023cd28d5f; harness 7de04fc11c6a692dfc5e1a86c7be817f3f0af8de.
36
+ See source for coverage.
37
+ - dataset:
38
+ id: llamaindex/ParseBench
39
+ task_id: text_content
40
+ revision: 2805a1d940f95a203e0ae4b88be9934f7765b3fc
41
+ value: 87.6431835862896
42
+ date: '2026-10-28'
43
+ source:
44
+ url: https://huggingface.co/KDLAI/KDL-Frontier-Parser-nano-v2/blob/5b5e0edda3c019563a2f6205ec49095288fafeb8/evaluation/parsebench-20261028-gpu5.json
45
+ name: Official ParseBench full-set author validation
46
+ user: Haigini
47
+ notes: Official full-set single pass; author-measured. Model 766505d578be5d99f2358ec9c95224023cd28d5f; harness 7de04fc11c6a692dfc5e1a86c7be817f3f0af8de.
48
+ See source for coverage.
49
+ - dataset:
50
+ id: llamaindex/ParseBench
51
+ task_id: text_formatting
52
+ revision: 2805a1d940f95a203e0ae4b88be9934f7765b3fc
53
+ value: 76.91130071725973
54
+ date: '2026-10-28'
55
+ source:
56
+ url: https://huggingface.co/KDLAI/KDL-Frontier-Parser-nano-v2/blob/5b5e0edda3c019563a2f6205ec49095288fafeb8/evaluation/parsebench-20261028-gpu5.json
57
+ name: Official ParseBench full-set author validation
58
+ user: Haigini
59
+ notes: Official full-set single pass; author-measured. Model 766505d578be5d99f2358ec9c95224023cd28d5f; harness 7de04fc11c6a692dfc5e1a86c7be817f3f0af8de.
60
+ See source for coverage.
61
+ - dataset:
62
+ id: llamaindex/ParseBench
63
+ task_id: layout
64
+ revision: 2805a1d940f95a203e0ae4b88be9934f7765b3fc
65
+ value: 75.31206769668802
66
+ date: '2026-10-28'
67
+ source:
68
+ url: https://huggingface.co/KDLAI/KDL-Frontier-Parser-nano-v2/blob/5b5e0edda3c019563a2f6205ec49095288fafeb8/evaluation/parsebench-20261028-gpu5.json
69
+ name: Official ParseBench full-set author validation
70
+ user: Haigini
71
+ notes: Official full-set single pass; author-measured. Model 766505d578be5d99f2358ec9c95224023cd28d5f; harness 7de04fc11c6a692dfc5e1a86c7be817f3f0af8de.
72
+ See source for coverage.