Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,72 @@
+---
+language: fa
+license: mit
+tags:
+- floret
+- fasttext
+- word-vectors
+- spacy
+- persian
+- farsi
+library_name: spacy
+---
+
+# fa-floret-wiki-vectors
+
+Floret word vectors for Persian (Farsi), trained on Persian Wikipedia with
+[floret-torch](https://github.com/Fazel94/floret-torch) — a GPU (PyTorch)
+implementation of [explosion/floret](https://github.com/explosion/floret),
+byte-exact hashing-compatible with CPU floret.
+
+## Corpus
+
+Persian Wikipedia (`fawiki` dump, WikiExtractor `--no-templates`, spaCy
+`blank("fa")` + sentencizer tokenization), full dump, no article cap:
+**8,428,449 sentences / 190,781,621 tokens**.
+
+## Training
+
+```
+python -m floret_torch.train \
+  --input fa.txt --output fa \
+  --model cbow --mode floret --dim 300 --minn 5 --maxn 5 \
+  --hashCount 2 --bucket 200000 --neg 10 --epoch 5 --lr 0.05 \
+  --minCount 20 --batch 8192 --dtype fp32 --device cuda --seed 0
+```
+
+`minn=maxn=5` (Persian is alphabetic/Perso-Arabic). `--bucket 200000` sizes
+the hashed vector table to 200k rows.
+
+## Files
+
+- `fa.floret` — the full hashed bucket table (`bucket dim minn maxn
+  hashCount hashSeed BOW EOW` header + rows). Reconstructs a vector for
+  *any* word, including ones unseen in training, via subword n-gram
+  hashing — this is the point of floret over plain word2vec, and matters
+  for Persian's rich morphology (clitics, inflection). Load with:
+  ```bash
+  python -m spacy init vectors fa fa.floret ./fa_pipeline --mode floret
+  ```
+- `fa.vec` — standard word2vec text format (`nwords dim` header, then
+  `word v0 .. vN` rows), the training vocabulary above `--minCount 20`.
+  Portable, no floret/spaCy dependency, but **no OOV coverage** — only
+  words present in this file have vectors. Load with gensim:
+  ```python
+  from gensim.models import KeyedVectors
+  kv = KeyedVectors.load_word2vec_format("fa.vec")
+  ```
+- `fa_floret-0.1.0-py3-none-any.whl` — installable spaCy pipeline package
+  (built via `spacy package --mode floret`) with the `.floret` table
+  embedded internally. Install and load directly:
+  ```bash
+  pip install fa_floret-0.1.0-py3-none-any.whl
+  ```
+  ```python
+  import fa_floret
+  nlp = fa_floret.load()
+  ```
+
+## License
+
+MIT, matching [floret-torch](https://github.com/Fazel94/floret-torch) and
+[explosion/floret](https://github.com/explosion/floret).