Update README.md

README.md

@@ -1,98 +1,79 @@
 ---
-license: cc-by-4.0
 language:
 - tr
-size_categories:
-- 10M<n<100M
+license: other
 task_categories:
 - text-generation
-tags:
-- pretraining
-- turkish
-- deduplication
-- quality-filtered
+arxiv: 2512.18834
 configs:
 - config_name: minhash_deduped
   data_files:
-    - split: train
-      path: "minhash_deduped/**/*.parquet"
-- config_name: quality_filtered
+  - split: train
+    path: minhash_deduped/**/*.parquet
+- config_name: matched
   data_files:
-    - split: train
-      path: "quality_filtered/**/*.parquet"
-- config_name: consensus
-  data_files:
-    - split: train
-      path: "consensus/*.parquet"
+  - split: train
+    path: consensus/*.parquet
+default: minhash_deduped
 ---
 
-# TurMix: Turkish Pretraining Data Mix
-
-A high-quality Turkish pretraining dataset created by combining, filtering, and deduplicating multiple sources.
-
-## Dataset Description
+<img src="https://huggingface.co/datasets/AdaMLLab/TurMix/resolve/main/finetasks_turkish_main_results.png" width="900" alt="Finetasks benchmark scores, showing TurMix-Matched as SOTA.">
 
-This dataset contains Turkish text from multiple web crawl sources, processed through a quality filtering and MinHash deduplication pipeline.
+TurMix ([https://arxiv.org/abs/2512.18834](https://arxiv.org/abs/2512.18834)) is a Turkish pretraining corpus containing 168 billion tokens across 219 million documents (in the minhash subset). Rather than scraping the web again, TurMix combines five publicly available Turkish datasets, applies Turkish-specific quality filtering, and performs cross-dataset deduplication.
 
-### Sources
-- **C4** (mC4 Turkish subset)
-- **CulturaX** (Turkish)
-- **Fineweb-2** (tur_Latn)
-- **HPLT-2** (tur_Latn, 5 shards)
-- **VNGRS Web Corpus**
+We train a 1.4B parameter language model through nanotron on 30 billion tokens to show that the `matched` subset of TurMix outperforms the previous state-of-the-art, [FineWeb-2 Turkish](https://huggingface.co/datasets/HuggingFaceFW/fineweb-2) (see [Appendix A9 in the Fineweb-2 paper](https://arxiv.org/pdf/2506.20920)), achieving a 5.5% relative improvement. Furthermore, the `minhash_deduped` subset performs competitively with over 2× the total number of tokens.
 
 ## Subsets
 
-### 1. `minhash_deduped` (Recommended)
-MinHash-deduplicated data. Each source was deduplicated individually to remove near-duplicate documents.
+| Subset | Documents | Tokens | Description |
+|--------|-----------|--------|-------------|
+| `minhash_deduped` | 219.1M | 167.6B | Document-level MinHash deduplication only |
+| `matched` | 67.6M | 56.0B | Documents appearing in 2+ source datasets |
 
-```python
-from datasets import load_dataset
-ds = load_dataset("AdaMLLab/TurMix", "minhash_deduped")
-```
+The matched subset uses cross-dataset agreement as a signal for quality.
 
-**Statistics:**
-- ~27M documents
-- 359GB compressed
-
-### 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/TurMix", "quality_filtered")
-```
 
-**Statistics:**
-- ~49M documents
-- 658GB 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/TurMix", "consensus")
+ds = load_dataset("AdaMLLab/TurMix", "minhash_deduped")
+ds = load_dataset("AdaMLLab/TurMix", "matched")
 ```
 
-**Statistics:**
-- 7.84M documents
-- 13GB compressed
-
-**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
-
-## Quality Filtering
+## Sources
+
+Tokens were counted using `meta-llama/Llama-3.2-3B`'s tokenizer.
+
+| Source | Tokens (MinHash) | Documents (MinHash) |
+|--------|------------------|---------------------|
+| HPLT 2.0 | 46.0B | 53.7M |
+| FineWeb-2 | 41.9B | 54.5M |
+| CulturaX | 35.8B | 47.9M |
+| C4 | 25.3B | 36.5M |
+| VNGRS-Web | 18.7B | 26.5M |
+| **Total** | **167.6B** | **219.1M** |
+
+## Pipeline
+
+1. Quality filtering with Turkish-specific thresholds (terminal punctuation, repetition patterns, Latin script ratio, 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
+
+```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}, 
+}
+```
 
-Documents were filtered based on:
-- Language identification (Turkish Latin script ratio)
-- Document length constraints
-- Line quality metrics
-- Repetition detection (including Turkish-specific patterns)
-- Boilerplate/policy phrase removal
+## License
 
-Filter thresholds based on Fineweb-2 Turkish configuration.
+See individual source dataset licenses.