Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,69 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
task_categories:
|
3 |
+
- question-answering
|
4 |
+
- table-question-answering
|
5 |
+
- text-generation
|
6 |
+
- text2text-generation
|
7 |
+
language:
|
8 |
+
- en
|
9 |
+
tags:
|
10 |
+
- text2sql
|
11 |
+
- text-to-sql
|
12 |
+
- database
|
13 |
+
- llm
|
14 |
+
- llama
|
15 |
+
pretty_name: birdbench
|
16 |
+
size_categories:
|
17 |
+
- 100M<n<1B
|
18 |
+
---
|
19 |
+
## BirdBench Dataset in DuckDB format
|
20 |
+
|
21 |
+
BirdBench is a benchmark for text-to-SQL capabilities, now available in DuckDB format for improved performance and usability.
|
22 |
+
|
23 |
+
## About BirdBench
|
24 |
+
|
25 |
+
BirdBench is a comprehensive benchmark dataset for evaluating text-to-SQL capabilities of language models. It features a diverse collection of databases spanning various domains including:
|
26 |
+
|
27 |
+
- Business and finance
|
28 |
+
- Entertainment and media
|
29 |
+
- Sports and recreation
|
30 |
+
- Health and medicine
|
31 |
+
- Education
|
32 |
+
- Travel and geography
|
33 |
+
- And many more
|
34 |
+
|
35 |
+
## Why DuckDB?
|
36 |
+
|
37 |
+
This repository contains the BirdBench dataset converted from SQLite to DuckDB format, which offers several advantages:
|
38 |
+
|
39 |
+
- **Performance**: DuckDB is significantly faster for analytical queries
|
40 |
+
- **Integration**: Better integration with Python data science tools
|
41 |
+
- **Features**: Support for vectorized operations and advanced analytical functions
|
42 |
+
- **Compatibility**: Works well in environments where SQLite might have limitations
|
43 |
+
|
44 |
+
## Dataset Structure
|
45 |
+
|
46 |
+
The dataset maintains the original BirdBench structure, with both training and validation databases converted to DuckDB format:
|
47 |
+
|
48 |
+
- `/train` - Contains training databases
|
49 |
+
- `/validation` - Contains validation databases
|
50 |
+
|
51 |
+
Each database preserves the original schema and data from the SQLite version.
|
52 |
+
|
53 |
+
## Usage
|
54 |
+
|
55 |
+
### Loading a database
|
56 |
+
|
57 |
+
```python
|
58 |
+
import duckdb
|
59 |
+
|
60 |
+
# Connect to a database
|
61 |
+
conn = duckdb.connect('path/to/database.duckdb')
|
62 |
+
|
63 |
+
# List tables
|
64 |
+
tables = conn.execute('SELECT name FROM sqlite_master WHERE type="table"').fetchall()
|
65 |
+
print(tables)
|
66 |
+
|
67 |
+
# Run a query
|
68 |
+
result = conn.execute('SELECT * FROM your_table LIMIT 5').fetchall()
|
69 |
+
print(result)
|