Update README.md

README.md

@@ -1,115 +1,86 @@
 ---
-license: cc-by-4.0
 language:
 - hi
-size_categories:
-- 10M<n<100M
+license: other
 task_categories:
 - text-generation
-tags:
-- pretraining
-- hindi
-- deduplication
-- quality-filtered
+arxiv: 2512.18834
 configs:
 - config_name: minhash_deduped
   data_files:
-    - split: train
-      path: "minhash_deduped/**/*.parquet"
+  - split: train
+    path: minhash_deduped/**/*.parquet
 - config_name: quality_filtered
   data_files:
-    - split: train
-      path: "quality_filtered/**/*.parquet"
-- config_name: consensus
+  - split: train
+    path: quality_filtered/**/*.parquet
+- config_name: matched
   data_files:
-    - split: train
-      path: "consensus/*.parquet"
+  - split: train
+    path: consensus/*.parquet
+default: minhash_deduped
 ---
 
-# HinMix: Hindi Pretraining Data Mix
+<img src="https://huggingface.co/datasets/AdaMLLab/HinMix/resolve/main/finetasks_hindi_main_results.png" width="900" alt="Finetasks benchmark scores, showing HinMix-MinHash as SOTA.">
 
-A high-quality Hindi pretraining dataset created by combining, filtering, and deduplicating multiple sources.
+HinMix ([https://arxiv.org/abs/2512.18834](https://arxiv.org/abs/2512.18834)) is a Hindi pretraining corpus containing 76 billion tokens across 60 million documents (in the minhash subset). Rather than scraping the web again, HinMix combines six publicly available Hindi datasets, applies Hindi-specific quality filtering, and performs cross-dataset deduplication.
 
-## Dataset Description
-
-This dataset contains Hindi text from multiple web crawl sources, processed through a quality filtering and MinHash deduplication pipeline.
-
-### Sources
-- **C4** (mC4 Hindi subset)
-- **CulturaX** (Hindi)
-- **Fineweb-2** (hin_Deva)
-- **HPLT-2** (hin_Deva)
-- **Sangraha** (verified and unverified Hindi splits)
+We train a 1.4B parameter language model through nanotron on 30 billion tokens to show that HinMix outperforms the previous state-of-the-art, [CulturaX Hindi](https://huggingface.co/datasets/uonlp/CulturaX) (see [Appendix A9 in the Fineweb-2 paper](https://arxiv.org/pdf/2506.20920)). The `minhash_deduped` subset achieves an 11.6% relative improvement, while the `matched` subset achieves an 8.1% relative improvement.
 
 ## Subsets
 
-### 1. `minhash_deduped` (Recommended)
-MinHash-deduplicated data. Each source was deduplicated individually to remove near-duplicate documents.
-
-```python
-from datasets import load_dataset
-ds = load_dataset("AdaMLLab/HinMix", "minhash_deduped")
-```
+| Subset | Documents | Tokens | Description |
+|--------|-----------|--------|-------------|
+| `quality_filtered` | 99.6M | 130.3B | Quality-filtered data before deduplication |
+| `minhash_deduped` | 59.6M | 76.2B | Document-level MinHash deduplication |
+| `matched` | 19.8M | 27.1B | Documents appearing in 2+ source datasets |
 
-**Statistics:**
-- ~60M documents
-- 136GB compressed
+The matched subset uses cross-dataset agreement as a signal for quality.
 
-### 2. `quality_filtered`
-Quality-filtered data before deduplication. Use this if you want to apply your own deduplication.
+## Usage
 
 ```python
 from datasets import load_dataset
-ds = load_dataset("AdaMLLab/HinMix", "quality_filtered")
-```
-
-**Statistics:**
-- ~99M documents
-- 231GB compressed
 
-### 3. `consensus`
-Documents that appear in 2+ sources (exact text match). These are high-confidence documents verified across multiple crawls.
-
-```python
-from datasets import load_dataset
-ds = load_dataset("AdaMLLab/HinMix", "consensus")
+ds = load_dataset("AdaMLLab/HinMix", "minhash_deduped")
+ds = load_dataset("AdaMLLab/HinMix", "quality_filtered")
+ds = load_dataset("AdaMLLab/HinMix", "matched")
 ```
 
-**Statistics:**
-- 1.92M documents
-- 3.7GB compressed
+## Sources
 
-**Schema:**
-- `text`: Document text
-- `id`: Primary document ID
-- `sources`: List of sources where document appears (e.g., `["c4", "culturax"]`)
-- `all_ids`: All document IDs from all sources
-- `metadata`: Additional metadata
+Tokens were counted using `meta-llama/Llama-3.2-3B`'s tokenizer.
 
-## Quality Filtering
+| Source | Tokens (MinHash) | Documents (MinHash) |
+|--------|------------------|---------------------|
+| FineWeb-2 | 20.0B | 17.1M |
+| CulturaX | 16.6B | 11.5M |
+| Sangraha (unverified) | 11.5B | 8.9M |
+| HPLT 2.0 | 10.2B | 6.7M |
+| Sangraha (verified) | 10.1B | 9.1M |
+| C4 | 7.7B | 6.3M |
+| **Total** | **76.2B** | **59.6M** |
 
-Documents were filtered based on:
-- Language identification (Hindi/Devanagari script ratio)
-- Document length constraints
-- Line quality metrics
-- Repetition detection
-- Boilerplate/policy phrase removal
+## Pipeline
 
-Filter thresholds based on Fineweb-2 Hindi configuration.
+1. Quality filtering with Hindi-specific thresholds (Devanagari script ratio, repetition patterns, language identification)
+2. Document-level MinHash deduplication (5-gram shingles, 14 bands, 8 hashes per band, similarity threshold 0.8)
+3. Cross-source matching to identify documents appearing in 2+ independent sources
 
 ## Citation
 
-If you use this dataset, please cite:
-
-```bibtex
-@dataset{hinmix2024,
-  title={HinMix: Hindi Pretraining Data Mix},
-  author={AdaMLLab},
-  year={2024},
-  publisher={Hugging Face}
+```bib
+@misc{alrashed2025mixminmatch,
+      title={Mix, MinHash, and Match: Cross-Source Agreement for Multilingual Pretraining Datasets}, 
+      author={Sultan Alrashed and Francesco Orabona},
+      year={2025},
+      eprint={2512.18834v2},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL},
+      url={https://arxiv.org/abs/2512.18834v2}, 
 }
 ```
 
 ## License
 
-This dataset is released under CC-BY-4.0. Individual source datasets may have their own licenses.
+See individual source dataset licenses.