sneakyfree commited on
Commit
2f398d9
·
verified ·
1 Parent(s): d989dc5

Add model card

Browse files
Files changed (1) hide show
  1. README.md +79 -0
README.md ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ tags:
4
+ - translation
5
+ - marian
6
+ - windyword
7
+ - west-slavic
8
+ - czech
9
+ - polish
10
+ - slovak
11
+ - english
12
+ language:
13
+ - zlw
14
+ - en
15
+ library_name: transformers
16
+ pipeline_tag: translation
17
+ ---
18
+
19
+ # WindyWord.ai Translation — West Slavic → English
20
+
21
+ **Translates West Slavic (Czech, Polish, Slovak) → English.**
22
+
23
+ **Quality Rating: ⭐⭐½ (2.5★ Basic)**
24
+
25
+ Part of the [WindyWord.ai](https://windyword.ai) translation fleet — 1,800+ proprietary language pairs.
26
+
27
+ ## Quality & Pricing Tier
28
+
29
+ - **5-star rating:** 2.5★ ⭐⭐½
30
+ - **Tier:** Basic
31
+ - **Composite score:** 51.3 / 100
32
+ - **Rated via:** Grand Rounds v2 — an 8-test stress battery (paragraphs, multi-paragraph, native input, domain stress, edge cases, round-trip fidelity, speed, and consistency checks)
33
+
34
+ ## Available Variants
35
+
36
+ This repository contains multiple deployment formats. Pick the one that matches your use case:
37
+
38
+ | Variant | Description |
39
+ |---|---|
40
+ | `lora/` | **WindyStandard** — our proprietary production baseline. Stable, reliable, optimized for GPU inference. |
41
+ | `lora-ct2-int8/` | **WindyStandard · CPU INT8** — CTranslate2 quantized version of WindyStandard. ~25% of the size, 2–4× faster on CPU, no measurable quality loss. |
42
+ | `herm0/` | **WindyEnhanced** — deep fine-tuned on OPUS-100, Tatoeba, and WikiMatrix parallel corpora. Measurably higher translation quality on supported pairs. |
43
+ | `herm0-ct2-int8/` | **WindyEnhanced · CPU INT8** — CTranslate2 quantized WindyEnhanced. Premium quality, CPU-efficient. |
44
+
45
+ ### Quick usage
46
+
47
+ **Transformers (PyTorch):**
48
+ ```python
49
+ from transformers import MarianMTModel, MarianTokenizer
50
+ tokenizer = MarianTokenizer.from_pretrained("WindyWord/translate-zlw-en", subfolder="lora")
51
+ model = MarianMTModel.from_pretrained("WindyWord/translate-zlw-en", subfolder="lora")
52
+ ```
53
+
54
+ **CTranslate2 (fast CPU inference):**
55
+ ```python
56
+ import ctranslate2
57
+ translator = ctranslate2.Translator("path/to/translate-zlw-en/lora-ct2-int8")
58
+ ```
59
+
60
+ ## Commercial Use
61
+
62
+ The WindyWord.ai platform provides:
63
+ - **Mobile apps** (iOS, Android — coming soon)
64
+ - **Real-time voice-to-text-to-translation** pipeline
65
+ - **API access** with premium model quality
66
+ - **Offline deployment** support
67
+
68
+ Visit [windyword.ai](https://windyword.ai) for apps and commercial API access.
69
+
70
+ ---
71
+
72
+ ## Provenance & License
73
+
74
+ Weights derived from the OPUS-MT project ([Helsinki-NLP/opus-mt-zlw-en](https://huggingface.co/Helsinki-NLP/opus-mt-zlw-en)) under CC-BY-4.0. WindyStandard, WindyEnhanced, and WindyScripture variants are proprietary to WindyWord.ai, independently trained and quality-certified via our Grand Rounds v2 test battery.
75
+
76
+ Licensed CC-BY-4.0 — attribution preserved as required.
77
+
78
+ *Certified by Opus 4.6 Opus-Claw (Dr. C) on Veron-1 (RTX 5090).*
79
+ *Patient file: [clinic record](https://github.com/sneakyfree/Windy-Clinic/blob/main/translation-pairs/zlw-en.json)*