Upload README.md with huggingface_hub

README.md

@@ -1,26 +1,80 @@
 ---
-dataset_info:
-  features:
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: paragraphs
-    sequence: string
-  splits:
-  - name: train
-    num_bytes: 14327395
-    num_examples: 4500
-  - name: validation
-    num_bytes: 1582995
-    num_examples: 500
-  download_size: 8846353
-  dataset_size: 15910390
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: validation
-    path: data/validation-*
+license: mit
+language:
+- de
+tags:
+- legal
+- german-law
+- question-answering
+- bgb
+- stgb
+size_categories:
+- 1K<n<10K
 ---
+
+# GerLayQA 5K Filtered Raw
+
+## Dataset Description
+
+This is a cleaned and filtered subset of the GerLayQA (German Legal Question Answering) dataset, specifically designed for German legal question-answering tasks. The dataset has been split into train and validation sets for proper model evaluation.
+
+## Dataset Statistics
+
+- **Total Samples**: 5000
+- **Train Samples**: 4500 (90.0%)
+- **Validation Samples**: 500 (10.0%)
+
+### Train Set
+- **Average Question Length**: 1100.3 characters (161.6 words)
+- **Average Answer Length**: 2002.2 characters (277.7 words)
+- **Unique Legal Paragraphs**: 1095
+
+### Validation Set
+- **Average Question Length**: 1084.1 characters (159.4 words)
+- **Average Answer Length**: 2001.7 characters (276.6 words)
+- **Unique Legal Paragraphs**: 339
+
+## Dataset Structure
+
+Each sample contains:
+- `question`: German legal question
+- `answer`: Detailed legal answer with citations
+- `paragraphs`: Relevant legal paragraphs (BGB, StGB, etc.)
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+# Load the dataset with splits
+dataset = load_dataset("DomainLLM/gerlayqa_5k_filtered_raw")
+
+# Access train and validation splits
+train_data = dataset['train']
+val_data = dataset['validation']
+
+# Example usage
+print(f"Train samples: {len(train_data)}")
+print(f"Validation samples: {len(val_data)}")
+print(f"Sample question: {train_data[0]['question']}")
+```
+
+## Cleaning Process
+
+This dataset has been cleaned using GPT-4o-mini to:
+- Remove HTML tags and formatting
+- Remove lawyer contact information and signatures
+- Normalize whitespace and punctuation
+- Preserve all legal content and citations
+
+## Split Information
+
+The dataset is split into:
+- **Train**: 4,500 samples (90%) for model training
+- **Validation**: 500 samples (10%) for model evaluation
+
+The split was created using a fixed random seed (42) to ensure reproducibility.
+
+## License
+
+This dataset is based on the original GerLayQA dataset and is released under the MIT license.