Add comprehensive README with tags and documentation

README.md

@@ -1,45 +1,19 @@
 ---
-dataset_info:
-- config_name: corpus
-  features:
-  - name: id
-    dtype: string
-  - name: text
-    dtype: string
-  - name: table_rows
-    dtype: int64
-  - name: table_cols
-    dtype: int64
-  - name: num_paragraphs
-    dtype: int64
-  splits:
-  - name: train
-    num_bytes: 889786
-    num_examples: 349
-  download_size: 436818
-  dataset_size: 889786
-- config_name: questions
-  features:
-  - name: id
-    dtype: string
-  - name: doc_id
-    dtype: string
-  - name: question
-    dtype: string
-  - name: answer
-    dtype: string
-  - name: answer_type
-    dtype: string
-  - name: answer_from
-    dtype: string
-  - name: chunk-must-contain
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 987525
-    num_examples: 2065
-  download_size: 298250
-  dataset_size: 987525
+license: mit
+task_categories:
+- question-answering
+- text-retrieval
+language:
+- en
+tags:
+- chunking
+- financial
+- tables
+- tat-qa
+- rag
+- retrieval
+size_categories:
+- 1K<n<10K
 configs:
 - config_name: corpus
   data_files:
@@ -50,3 +24,90 @@ configs:
   - split: train
     path: questions/train-*
 ---
+
+<div align="center">
+
+# 📊 Tacha: Table Chunking Assessment
+
+**Ta**ble **Ch**unking **A**ssessment - A dataset for evaluating chunking algorithms on financial documents with tables.
+
+</div>
+
+## Overview
+
+Tacha is designed to test how well chunking algorithms handle **structured tabular data** within documents. It contains financial reports with markdown tables and accompanying text, derived from the [TAT-QA](https://nextplusplus.github.io/TAT-QA/) dataset.
+
+### Key Challenges
+
+- Keeping table headers with data rows
+- Not splitting tables mid-row
+- Preserving table-text relationships
+- Handling multi-column financial data
+
+## Dataset Structure
+
+### Corpus
+
+The `corpus` config contains 349 financial documents with tables (12+ rows, 4+ columns).
+
+| Column | Type | Description |
+|--------|------|-------------|
+| `id` | string | Unique document identifier |
+| `text` | string | Full document with markdown tables and text |
+| `table_rows` | int | Number of rows in the main table |
+| `table_cols` | int | Number of columns in the main table |
+| `num_paragraphs` | int | Number of text paragraphs accompanying the table |
+
+### Questions
+
+The `questions` config contains 2,065 questions requiring understanding of table content.
+
+| Column | Type | Description |
+|--------|------|-------------|
+| `id` | string | Unique question identifier |
+| `doc_id` | string | Reference to corpus document |
+| `question` | string | Question about the table/document content |
+| `answer` | string | Answer to the question |
+| `answer_type` | string | Type: span, multi-span, arithmetic, count |
+| `answer_from` | string | Source: table, text, or table-text |
+| `chunk-must-contain` | string | Passage that a good chunk must contain |
+
+## Statistics
+
+| Metric | Value |
+|--------|-------|
+| Documents | 349 |
+| Questions | 2,065 |
+| Avg table rows | 15.8 |
+| Avg table cols | 4.6 |
+| Answer types | 42% arithmetic, 42% span, 13% multi-span, 2% count |
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+# Load the corpus
+corpus = load_dataset("chonkie-ai/tacha", "corpus", split="train")
+
+# Load the questions
+questions = load_dataset("chonkie-ai/tacha", "questions", split="train")
+
+# Use with MTCB evaluator
+from mtcb import TachaEvaluator
+from chonkie import RecursiveChunker
+
+evaluator = TachaEvaluator(
+    chunker=RecursiveChunker(chunk_size=512),
+    embedding_model="voyage-3-large"
+)
+result = evaluator.evaluate(k=[1, 3, 5, 10])
+```
+
+## Source
+
+Derived from [TAT-QA](https://nextplusplus.github.io/TAT-QA/) (ACL 2021), filtered to documents with larger tables (12+ rows, 4+ columns).
+
+## License
+
+MIT License