---
extra_gated_heading: "Access Request for Malayalam Books Dataset"
extra_gated_prompt: >
  To access this dataset, please provide the following information. 
  You agree to use this dataset for non-commercial purposes only and not to conduct experiments that cause harm to human subjects.
  These brief questions help us ensure access goes to genuine researchers or developers with clear intent—making our approval process more secure and fair.
extra_gated_description: >
  It may take maximum 24hours to process your request.
extra_gated_button_content: "Acknowledge License"
extra_gated_fields:
  Full_Name: text
  email_address(please dont use disposable mail ids, your requests wont be considered): text
  intended_use:
    type: select
    options:
      - Research
      - Personal education
      - { label: "Other", value: other }
  acknowledgment_checkbox:
    type: checkbox
    label: "I agree not to redistribute this dataset"
---
---
pretty_name: Malayalam Books from Granthappura (File Dataset)
language:
  - ml
tags:
  - malayalam
  - books
  - literature
  - kerala
  - digital-archive
  - historical-texts
  - pdf
  - scanned
license: cc-by-4.0
size_categories:
  - 1K<n<10K
# This card documents a file-based dataset (PDFs in LFS + compact index)
---

# Malayalam Books from Granthappura Digital Archive

⚠️ Important Notice

- Full credit to Granthappura Digital Archive (https://gpura.org). This collection mirrors their publicly available scans for research/education.
- Content removal policy: If Granthappura or any rightsholder requests removal, the dataset will be taken down immediately.
- Rights: The dataset card and index are shared under CC BY 4.0; PDF content rights remain with Granthappura and/or original publishers/authors.

## Overview

Historical Malayalam books and documents from the Granthappura Digital Archive. Items span the late 1800s–1900s and include classical literature, religious works, cinema song books, and other culturally significant texts.

This repository is a file-based dataset:
- PDFs are stored directly in the repo (Git LFS).
- A compact index (CSV and Parquet) provides metadata and paths.
- No PDF bytes are embedded inside Parquet to avoid massive shards.

Estimated size: ~89 GB of PDFs (will vary as the collection grows). Number of items is in the low-thousands.

## Repository Layout

- `pdfs/` — All PDF files (Git LFS)
- `index/index.csv` — Compact index (CSV)
- `index/index.parquet` — Compact index (Parquet)
- `README.md` — This dataset card

### Index Schema

Each row corresponds to one PDF:
- `filename`: PDF file name
- `path_in_repo`: Relative path to the file in this repo (e.g., `pdfs/<file>.pdf`)
- `size_bytes`: File size in bytes
- `title`: Title parsed from filename (best-effort)
- `year`: Publication year (best-effort)
- `author`: Author (if parsable)
- `publisher`: Usually "Granthappura"
- `language`: Malayalam (ml)
- `type`: e.g., `book`

Notes:
- Metadata is extracted heuristically from filenames and may be incomplete or noisy.
- Users should validate/augment metadata as needed for their research.

## How to Use

### Option A: Download index + PDFs locally

```python
from pathlib import Path
from huggingface_hub import snapshot_download
import pandas as pd

repo_id = "cazzz307/Malayalam-books-granthappura-new"
local_dir = snapshot_download(
    repo_id=repo_id,
    repo_type="dataset",
    allow_patterns=["index/*", "pdfs/*.pdf"], 
)

index_path = Path(local_dir) / "index/index.parquet"
df = pd.read_parquet(index_path)
print(len(df), "rows")


pdf_path = Path(local_dir) / df.loc[0, "path_in_repo"]
print(pdf_path.exists())
```

This will download many GBs. Ensure sufficient disk and bandwidth.

### Option B: Read the index only (remote)

```python
import pandas as pd
repo_id = "cazzz307/Malayalam-books-granthappura-new"
base = f"https://huggingface.co/datasets/{repo_id}/resolve/main/index"
index_url = f"{base}/index.csv"  

df = pd.read_csv(index_url)
print(df.head())


import requests
pdf_url = f"https://huggingface.co/datasets/{repo_id}/resolve/main/" + df.loc[0, "path_in_repo"]
r = requests.get(pdf_url)
open("sample.pdf", "wb").write(r.content)
```

### Use Cases
- OCR benchmarking and training (Malayalam)
- Historical text analysis / digital humanities
- Document layout/vision tasks on scanned material
- Retrieval and indexing research on PDFs

## Ethical Use and Rights

- Attribution: Please cite and credit the original source — Granthappura Digital Archive (https://gpura.org).
- Rights: The PDFs are historical scans from the archive; rights remain with Granthappura and/or original publishers.
- License: This dataset’s index and card are under CC BY 4.0. Do not assume this license applies to the PDF contents.
- Removal: Contact via the dataset page for immediate takedown requests.

## Citation

Primary attribution:
```
Original Source: Granthappura Digital Archive (https://gpura.org)
Kerala Digital Heritage Collection
```

Dataset citation:
```
Malayalam Historical Books (File Dataset)
Aggregated from: Granthappura Digital Archive (https://gpura.org)
Accessed: August 2025
Note: All rights remain with original source and rightsholders.
```

## Limitations
- Metadata is best-effort and may be incomplete.
- PDFs are large; downloading the entire set requires significant bandwidth and storage.
- No OCR text is provided in this dataset; PDFs are scans.

## Changelog
- 2025-08: Initial release as a file-based dataset with compact index (no PDFs embedded in Parquet).