failed09 commited on
Commit
dce2032
·
verified ·
1 Parent(s): 513accb

Card: use generic source wording

Browse files
Files changed (2) hide show
  1. README.md +5 -5
  2. SHA256SUMS +1 -1
README.md CHANGED
@@ -22,8 +22,8 @@ tags:
22
 
23
  ## Overview
24
 
25
- Compact subword word embeddings for Bashkir (`ba`), trained on a cleaned
26
- monolingual collection of Bashkir-language text. The release ships a single
27
  quantized model (`bashkir_fasttext_300d_mini.bin`) intended for semantic search,
28
  lexical tools and lightweight NLP experiments on CPU, mobile and edge devices.
29
 
@@ -31,7 +31,7 @@ lexical tools and lightweight NLP experiments on CPU, mobile and edge devices.
31
  | --- | --- |
32
  | Task | Feature extraction / word embeddings |
33
  | Default artifact | `bashkir_fasttext_300d_mini.bin` |
34
- | Source | A monolingual Bashkir-language text collection |
35
  | Version / license | v0.1-beta / Apache-2.0 |
36
 
37
  ## Contents
@@ -75,8 +75,8 @@ monolingual Bashkir text → Unicode NFC tokenization → skip-gram 300d (char n
75
  → prune (top-50k words, top-100k subwords) → PQ FP16 → mini .bin
76
  ```
77
 
78
- The model was trained from a cleaned monolingual collection of Bashkir-language
79
- text. Corpus size, source reference and training parameters are recorded in
80
  [`config.json`](config.json). Text preparation uses normalized Unicode tokens and
81
  preserves Bashkir-specific letters. This release is a compact derivative of the
82
  full 300-dimensional model: it retains subword information while reducing the
 
22
 
23
  ## Overview
24
 
25
+ Compact subword word embeddings for Bashkir (`ba`), trained on a monolingual
26
+ Bashkir-language dataset. The release ships a single
27
  quantized model (`bashkir_fasttext_300d_mini.bin`) intended for semantic search,
28
  lexical tools and lightweight NLP experiments on CPU, mobile and edge devices.
29
 
 
31
  | --- | --- |
32
  | Task | Feature extraction / word embeddings |
33
  | Default artifact | `bashkir_fasttext_300d_mini.bin` |
34
+ | Source | A monolingual Bashkir-language dataset |
35
  | Version / license | v0.1-beta / Apache-2.0 |
36
 
37
  ## Contents
 
75
  → prune (top-50k words, top-100k subwords) → PQ FP16 → mini .bin
76
  ```
77
 
78
+ The model was trained from a monolingual Bashkir-language dataset.
79
+ Corpus size, source reference and training parameters are recorded in
80
  [`config.json`](config.json). Text preparation uses normalized Unicode tokens and
81
  preserves Bashkir-specific letters. This release is a compact derivative of the
82
  full 300-dimensional model: it retains subword information while reducing the
SHA256SUMS CHANGED
@@ -1,4 +1,4 @@
1
  23839280d4c0e4a3c7392795e4b8ec7487b219c5a7cc8c6063802f92ec120e4f META.json
2
- 3e0200faed8704a6aa69850fafb5c0ff720bbe891c0a2f1beb0080d1c3044a1a README.md
3
  51c65280ece23c79da7fc3f655da7eb9bbee3b8dd3684086d58e8ddabe0096e6 bashkir_fasttext_300d_mini.bin
4
  7d94cf89587e637d4df24a34a4d9608412a9e77d7e99e50088e53783b781d37d config.json
 
1
  23839280d4c0e4a3c7392795e4b8ec7487b219c5a7cc8c6063802f92ec120e4f META.json
2
+ 434bbbcad59f1c6bf946c331c59bca3370a9e29ac1deb79e35225e0a900fe539 README.md
3
  51c65280ece23c79da7fc3f655da7eb9bbee3b8dd3684086d58e8ddabe0096e6 bashkir_fasttext_300d_mini.bin
4
  7d94cf89587e637d4df24a34a4d9608412a9e77d7e99e50088e53783b781d37d config.json