Register ExtractBench Pro as a benchmark and restore RL discovery
#4
by xujustinj - opened
README.md
CHANGED
|
@@ -38,6 +38,7 @@ pretty_name: ExtractBench Pro
|
|
| 38 |
size_categories:
|
| 39 |
- n<1K
|
| 40 |
tags:
|
|
|
|
| 41 |
- harbor
|
| 42 |
- document-extraction
|
| 43 |
- structured-extraction
|
|
|
|
| 38 |
size_categories:
|
| 39 |
- n<1K
|
| 40 |
tags:
|
| 41 |
+
- rl-environment
|
| 42 |
- harbor
|
| 43 |
- document-extraction
|
| 44 |
- structured-extraction
|
eval.yaml
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
name: ExtractBench Pro
|
| 2 |
+
description: ExtractBench Pro evaluates schema-guided extraction across six challenging
|
| 3 |
+
document categories.
|
| 4 |
+
evaluation_framework: extractbench
|
| 5 |
+
tasks:
|
| 6 |
+
- id: mean
|
| 7 |
+
config: extract-bench-pro
|
| 8 |
+
split: mean
|
| 9 |
+
- id: document_conventions
|
| 10 |
+
config: extract-bench-pro
|
| 11 |
+
split: document_conventions
|
| 12 |
+
- id: house_vocabulary
|
| 13 |
+
config: extract-bench-pro
|
| 14 |
+
split: house_vocabulary
|
| 15 |
+
- id: house_counting_rules
|
| 16 |
+
config: extract-bench-pro
|
| 17 |
+
split: house_counting_rules
|
| 18 |
+
- id: drawn_relations
|
| 19 |
+
config: extract-bench-pro
|
| 20 |
+
split: drawn_relations
|
| 21 |
+
- id: charts_inside_documents
|
| 22 |
+
config: extract-bench-pro
|
| 23 |
+
split: charts_inside_documents
|
| 24 |
+
- id: long_chart_sheets
|
| 25 |
+
config: extract-bench-pro
|
| 26 |
+
split: long_chart_sheets
|