sneakyfree commited on
Commit
4dde596
·
verified ·
1 Parent(s): 5dcce46

clone from WindyWord/translate-zls-zls via local upload (ADR-039 Phase C)

Browse files
.gitattributes CHANGED
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ lora/source.spm filter=lfs diff=lfs merge=lfs -text
37
+ lora/target.spm filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ tags:
4
+ - translation
5
+ - marian
6
+ - windyword
7
+ - south-slavic
8
+ - bulgarian
9
+ - croatian
10
+ - macedonian
11
+ - serbian
12
+ - slovenian
13
+ language:
14
+ - zls
15
+ library_name: transformers
16
+ pipeline_tag: translation
17
+ ---
18
+
19
+ # WindyWord.ai Translation — South Slavic → South Slavic
20
+
21
+ **Translates South Slavic (Bulgarian, Croatian, Macedonian, Serbian, Slovenian) → South Slavic (Bulgarian, Croatian, Macedonian, Serbian, Slovenian).**
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:** 53.1 / 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
+
43
+ ### Quick usage
44
+
45
+ **Transformers (PyTorch):**
46
+ ```python
47
+ from transformers import MarianMTModel, MarianTokenizer
48
+ tokenizer = MarianTokenizer.from_pretrained("WindyWord/translate-zls-zls", subfolder="lora")
49
+ model = MarianMTModel.from_pretrained("WindyWord/translate-zls-zls", subfolder="lora")
50
+ ```
51
+
52
+ **CTranslate2 (fast CPU inference):**
53
+ ```python
54
+ import ctranslate2
55
+ translator = ctranslate2.Translator("path/to/translate-zls-zls/lora-ct2-int8")
56
+ ```
57
+
58
+ ## Commercial Use
59
+
60
+ The WindyWord.ai platform provides:
61
+ - **Mobile apps** (iOS, Android — coming soon)
62
+ - **Real-time voice-to-text-to-translation** pipeline
63
+ - **API access** with premium model quality
64
+ - **Offline deployment** support
65
+
66
+ Visit [windyword.ai](https://windyword.ai) for apps and commercial API access.
67
+
68
+ ---
69
+
70
+ ## Provenance & License
71
+
72
+ Weights derived from the OPUS-MT project ([Helsinki-NLP/opus-mt-zls-zls](https://huggingface.co/Helsinki-NLP/opus-mt-zls-zls)) 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.
73
+
74
+ Licensed CC-BY-4.0 — attribution preserved as required.
75
+
76
+ *Certified by Opus 4.6 Opus-Claw (Dr. C) on Veron-1 (RTX 5090).*
77
+ *Patient file: [clinic record](https://github.com/sneakyfree/Windy-Clinic/blob/main/translation-pairs/zls-zls.json)*
lora-ct2-int8/config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_source_bos": false,
3
+ "add_source_eos": false,
4
+ "bos_token": "<s>",
5
+ "decoder_start_token": "</s>",
6
+ "eos_token": "</s>",
7
+ "layer_norm_epsilon": null,
8
+ "multi_query_attention": false,
9
+ "unk_token": "<unk>"
10
+ }
lora-ct2-int8/model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89c5bff9c6f77c7bdcfe4207cafff3425dd45b194447921155627cca2ad81620
3
+ size 63134595
lora-ct2-int8/shared_vocabulary.json ADDED
The diff for this file is too large to render. See raw diff
 
lora/config.json ADDED
@@ -0,0 +1,56 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "Helsinki-NLP/opus-mt-zls-zls",
3
+ "activation_dropout": 0.0,
4
+ "activation_function": "swish",
5
+ "add_bias_logits": false,
6
+ "add_final_layer_norm": false,
7
+ "architectures": [
8
+ "MarianMTModel"
9
+ ],
10
+ "attention_dropout": 0.0,
11
+ "bos_token_id": 0,
12
+ "classif_dropout": 0.0,
13
+ "classifier_dropout": 0.0,
14
+ "d_model": 512,
15
+ "decoder_attention_heads": 8,
16
+ "decoder_ffn_dim": 2048,
17
+ "decoder_layerdrop": 0.0,
18
+ "decoder_layers": 6,
19
+ "decoder_start_token_id": 33398,
20
+ "decoder_vocab_size": 33399,
21
+ "dropout": 0.1,
22
+ "encoder_attention_heads": 8,
23
+ "encoder_ffn_dim": 2048,
24
+ "encoder_layerdrop": 0.0,
25
+ "encoder_layers": 6,
26
+ "eos_token_id": 0,
27
+ "extra_pos_embeddings": 33399,
28
+ "forced_eos_token_id": 0,
29
+ "id2label": {
30
+ "0": "LABEL_0",
31
+ "1": "LABEL_1",
32
+ "2": "LABEL_2"
33
+ },
34
+ "init_std": 0.02,
35
+ "is_encoder_decoder": true,
36
+ "label2id": {
37
+ "LABEL_0": 0,
38
+ "LABEL_1": 1,
39
+ "LABEL_2": 2
40
+ },
41
+ "max_length": null,
42
+ "max_position_embeddings": 512,
43
+ "model_type": "marian",
44
+ "normalize_before": false,
45
+ "normalize_embedding": false,
46
+ "num_beams": null,
47
+ "num_hidden_layers": 6,
48
+ "pad_token_id": 33398,
49
+ "scale_embedding": true,
50
+ "share_encoder_decoder_embeddings": true,
51
+ "static_position_embeddings": true,
52
+ "torch_dtype": "float32",
53
+ "transformers_version": "4.45.1",
54
+ "use_cache": true,
55
+ "vocab_size": 33399
56
+ }
lora/generation_config.json ADDED
@@ -0,0 +1,16 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bad_words_ids": [
3
+ [
4
+ 33398
5
+ ]
6
+ ],
7
+ "bos_token_id": 0,
8
+ "decoder_start_token_id": 33398,
9
+ "eos_token_id": 0,
10
+ "forced_eos_token_id": 0,
11
+ "max_length": 512,
12
+ "num_beams": 6,
13
+ "pad_token_id": 33398,
14
+ "renormalize_logits": true,
15
+ "transformers_version": "4.45.1"
16
+ }
lora/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ee368a092d12c2b4687bc59310f764bf2afef387135bb9e50e6c7574bd28499c
3
+ size 245117788
lora/source.spm ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bd08ddd34f5d0f26a28e54a4ab90b4179ca0708c701987fe84f8bab7d0b25654
3
+ size 988338
lora/special_tokens_map.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "eos_token": "</s>",
3
+ "pad_token": "<pad>",
4
+ "unk_token": "<unk>"
5
+ }
lora/target.spm ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:85cd24766aa9a7e50f6be6ed48564fc600f8667ee07919cdb6633e9677f2b8fc
3
+ size 988452
lora/tokenizer_config.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "added_tokens_decoder": {
3
+ "0": {
4
+ "content": "</s>",
5
+ "lstrip": false,
6
+ "normalized": false,
7
+ "rstrip": false,
8
+ "single_word": false,
9
+ "special": true
10
+ },
11
+ "1": {
12
+ "content": "<unk>",
13
+ "lstrip": false,
14
+ "normalized": false,
15
+ "rstrip": false,
16
+ "single_word": false,
17
+ "special": true
18
+ },
19
+ "33398": {
20
+ "content": "<pad>",
21
+ "lstrip": false,
22
+ "normalized": false,
23
+ "rstrip": false,
24
+ "single_word": false,
25
+ "special": true
26
+ }
27
+ },
28
+ "clean_up_tokenization_spaces": false,
29
+ "eos_token": "</s>",
30
+ "model_max_length": 512,
31
+ "pad_token": "<pad>",
32
+ "separate_vocabs": false,
33
+ "source_lang": "zls",
34
+ "sp_model_kwargs": {},
35
+ "target_lang": "zls",
36
+ "tokenizer_class": "MarianTokenizer",
37
+ "unk_token": "<unk>"
38
+ }
lora/vocab.json ADDED
The diff for this file is too large to render. See raw diff