Upload folder using huggingface_hub

README.md

@@ -11,39 +11,17 @@ A multilingual subset of [allenai/c4](https://huggingface.co/datasets/allenai/c4
 
 ## Language Composition
 
-| Language | Code | Family | Samples | Tokens | Avg tok/doc | % of tokens |
-|----------|------|--------|--------:|-------:|------------:|------------:|
-| English | `en` | Germanic | 1,440 | 707,314 | 491.2 | 35.8% |
-| Chinese | `zh` | Sino-Tibetan | 273 | 232,924 | 853.2 | 11.8% |
-| Japanese | `ja` | Japonic | 73 | 143,041 | 1,959.5 | 7.2% |
-| Spanish | `es` | Romance | 152 | 149,495 | 983.5 | 7.6% |
-| French | `fr` | Romance | 93 | 120,373 | 1,294.3 | 6.1% |
-| German | `de` | Germanic | 130 | 95,043 | 731.1 | 4.8% |
-| Korean | `ko` | Koreanic | 44 | 67,377 | 1,531.3 | 3.4% |
-| Russian | `ru` | Slavic | 119 | 104,807 | 880.7 | 5.3% |
-| Arabic | `ar` | Semitic | 96 | 115,058 | 1,198.5 | 5.8% |
-| Portuguese | `pt` | Romance | 129 | 78,621 | 609.5 | 4.0% |
-| Hindi | `hi` | Indo-Aryan | 60 | 65,634 | 1,093.9 | 3.3% |
-| Tamil | `ta` | Dravidian | 57 | 98,417 | 1,726.6 | 5.0% |
-
-## Design Principles
-
-- **~2M token budget** calibrated for stable perplexity convergence
-- **Language distribution** weighted by LLM usage patterns, global speaker base, and language family diversity
-- **9 language families** across 12 languages to maximize typological coverage
-- **8 distinct scripts:** Latin, CJK, Kana, Hangul, Cyrillic, Arabic, Devanagari, Tamil
-- Similar language families are capped to avoid over-representation (e.g., 3 Romance languages share ~18% instead of dominating)
-
-## Usage
-
-```python
-import pandas as pd
-
-df = pd.read_parquet("data/c4_evaluation/c4_evaluation.parquet")
-
-# Filter by language
-en_texts = df[df["language"] == "en"]["text"].tolist()
-
-# Get all languages
-print(df["language"].value_counts())
-```
+| Language | Code | Family | Samples | Tokens | % of tokens |
+|----------|------|--------|--------:|-------:|------------:|
+| English | `en` | Germanic | 1,440 | 707,314 | 35.8% |
+| Chinese | `zh` | Sino-Tibetan | 273 | 232,924 | 11.8% |
+| Japanese | `ja` | Japonic | 73 | 143,041 | 7.2% |
+| Spanish | `es` | Romance | 152 | 149,495 | 7.6% |
+| French | `fr` | Romance | 93 | 120,373 | 6.1% |
+| German | `de` | Germanic | 130 | 95,043 | 4.8% |
+| Korean | `ko` | Koreanic | 44 | 67,377 | 3.4% |
+| Russian | `ru` | Slavic | 119 | 104,807 | 5.3% |
+| Arabic | `ar` | Semitic | 96 | 115,058 | 5.8% |
+| Portuguese | `pt` | Romance | 129 | 78,621 | 4.0% |
+| Hindi | `hi` | Indo-Aryan | 60 | 65,634 | 3.3% |
+| Tamil | `ta` | Dravidian | 57 | 98,417 | 5.0% |