Upload README.md with huggingface_hub

README.md

@@ -1,3 +1,72 @@
 ---
-license: mit
+license: other
+language:
+- km
+- en
+task_categories:
+- text-generation
+- fill-mask
+tags:
+- khmer
+- cambodia
+- code-switching
+- sentencepiece
+- diffusion-lm
+pretty_name: Khmer + English Mixed Text Corpus
 ---
+
+# Khmer + English Mixed Text Corpus
+
+A cleaned, deduplicated Khmer text corpus with naturally-occurring **Khmer/English code-switching**
+(English tech & finance terms, Latin script, and digits embedded in Khmer text). It is the training
+data for the [`Panhapich/khmer-sp-8k`](https://huggingface.co/Panhapich/khmer-sp-8k) SentencePiece tokenizer
+and the text-only warm-start of a shared Khmer diffusion decoder.
+
+## Contents
+
+| File | Description |
+|---|---|
+| `all_text.txt` | The corpus. One normalized sentence per line, UTF-8. |
+| `metadata.json` | Corpus statistics and provenance. |
+
+- **Total sentences:** 1,959,780
+
+## Sources
+
+Built from exactly four sources, mixed together and shuffled. A global cap
+(`2,000,000` sentences) bounds the total — small sources and the local file are taken
+in full, and the large raw-text source fills the remainder:
+
+| Source | Field | Contribution |
+|---|---|---|
+| `Sokheng/khmer-synthetic-ocr-v1-100k` | `text` | Synthetic OCR receipt/label text (Khmer/English/digit mixed) |
+| `rinabuoy/aupp-assignment-data` | `title` | Khmer news headlines |
+| `nphearum/khmer-raw-text-3M-v2` | `text` | Large raw Khmer text (split on the Khmer full stop `។`) |
+| `khmer_corpus.txt` (local) | line | Pre-chunked ~1000-character Khmer text blocks |
+
+**The Khmer/English mixing is entirely organic** — it comes from the real sources. No synthetic
+code-switching is injected.
+
+## Preprocessing
+
+Every sentence is normalized: coerce to string, strip, replace newlines/tabs with a single space,
+collapse repeated whitespace, and drop empty/invalid labels (`???`, `NULL`, `N/A`, `UNKNOWN`).
+Khmer word-spacing, punctuation, Khmer + Arabic numerals, and English/Latin tokens are preserved.
+
+Hub-source sentences are kept only if 5-400 characters; the local blocks are exempt from the length
+cap. The corpus is deduplicated and shuffled with a fixed seed (42) for reproducibility.
+
+## Intended use
+
+Subword tokenizer training for Khmer with code-switched English, text-only diffusion/masked-LM
+pretraining of a Khmer decoder, and general mixed-script Khmer NLP.
+
+## Out of scope / limitations
+
+Plain text only (no labels or translations). Includes synthetic OCR text and raw web-scraped text,
+so expect some noise and non-standard spelling. Not filtered for PII beyond the rules above.
+
+## Licensing
+
+This corpus is derived from third-party datasets and inherits their licenses. Confirm the terms of
+each source above and set the `license` field accordingly — `other` is a placeholder.