File size: 1,408 Bytes
ed4316d
45e3732
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ed4316d
45e3732
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
---
license: apache-2.0
task_categories:
- text-generation
- text-classification
language:
- en
- hi
tags:
- invoice-extraction
- document-ai
- pydantic-schema
- finetune
- chatml
size_categories:
- 1K<n<10K
---

# MindMap Enterprise Invoice Extraction & Fine-Tuning Dataset

Production-grade benchmark and fine-tuning dataset built from **1,007 enterprise PDF invoices** for extracting structured JSON metadata matching strict target Pydantic schemas.

## Dataset Structure & Splits

- `train` (1,074 examples): 80% training split containing English and Hindi Devanagari invoice document pairs.
- `validation` (134 examples): 10% validation split.
- `test_golden` (135 examples): 10% held-out test split for automated F1 and schema validity scoring.
- `catastrophic_forgetting_benchmark` (5 examples): General capability benchmark suite (Math, Coding, Reasoning).

## Target Schema Specifications

```json
{
  "invoice_number": "string",
  "vendor_name": "string",
  "invoice_date": "YYYY-MM-DD",
  "line_items": [
    {
      "description": "string",
      "quantity": "number",
      "unit_price": "number"
    }
  ],
  "subtotal": "number",
  "tax_amount": "number",
  "total_amount": "number",
  "currency": "string (ISO 4217)"
}
```

## Quick Usage with Hugging Face Datasets

```python
from datasets import load_dataset

ds = load_dataset("Msduck/invoice_dataset")
print(ds)
print(ds["train"][0])
```