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

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+lora/source.spm filter=lfs diff=lfs merge=lfs -text
+lora/target.spm filter=lfs diff=lfs merge=lfs -text
+herm0/source.spm filter=lfs diff=lfs merge=lfs -text
+herm0/target.spm filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,80 @@
+---
+license: cc-by-4.0
+tags:
+- translation
+- marian
+- windyword
+- cushitic-languages
+- somali
+- oromo
+- afar
+- beja
+- english
+language:
+- cus
+- en
+library_name: transformers
+pipeline_tag: translation
+---
+
+# WindyWord.ai Translation — Cushitic languages → English
+
+**Translates Cushitic languages (Somali, Oromo, Afar, Beja) → English.**
+
+**Quality Rating: ⭐⭐½  (2.5★ Basic)**
+
+Part of the [WindyWord.ai](https://windyword.ai) translation fleet — 1,800+ proprietary language pairs.
+
+## Quality & Pricing Tier
+
+- **5-star rating:** 2.5★ ⭐⭐½
+- **Tier:** Basic
+- **Composite score:** 51.0 / 100
+- **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)
+
+## Available Variants
+
+This repository contains multiple deployment formats. Pick the one that matches your use case:
+
+| Variant | Description |
+|---|---|
+| `lora/` | **WindyStandard** — our proprietary production baseline. Stable, reliable, optimized for GPU inference. |
+| `lora-ct2-int8/` | **WindyStandard · CPU INT8** — CTranslate2 quantized version of WindyStandard. ~25% of the size, 2–4× faster on CPU, no measurable quality loss. |
+| `herm0/` | **WindyEnhanced** — deep fine-tuned on OPUS-100, Tatoeba, and WikiMatrix parallel corpora. Measurably higher translation quality on supported pairs. |
+| `herm0-ct2-int8/` | **WindyEnhanced · CPU INT8** — CTranslate2 quantized WindyEnhanced. Premium quality, CPU-efficient. |
+
+### Quick usage
+
+**Transformers (PyTorch):**
+```python
+from transformers import MarianMTModel, MarianTokenizer
+tokenizer = MarianTokenizer.from_pretrained("WindyWord/translate-cus-en", subfolder="lora")
+model = MarianMTModel.from_pretrained("WindyWord/translate-cus-en", subfolder="lora")
+```
+
+**CTranslate2 (fast CPU inference):**
+```python
+import ctranslate2
+translator = ctranslate2.Translator("path/to/translate-cus-en/lora-ct2-int8")
+```
+
+## Commercial Use
+
+The WindyWord.ai platform provides:
+- **Mobile apps** (iOS, Android — coming soon)
+- **Real-time voice-to-text-to-translation** pipeline
+- **API access** with premium model quality
+- **Offline deployment** support
+
+Visit [windyword.ai](https://windyword.ai) for apps and commercial API access.
+
+---
+
+## Provenance & License
+
+Weights derived from the OPUS-MT project ([Helsinki-NLP/opus-mt-cus-en](https://huggingface.co/Helsinki-NLP/opus-mt-cus-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.
+
+Licensed CC-BY-4.0 — attribution preserved as required.
+
+*Certified by Opus 4.6 Opus-Claw (Dr. C) on Veron-1 (RTX 5090).*
+*Patient file: [clinic record](https://github.com/sneakyfree/Windy-Clinic/blob/main/translation-pairs/cus-en.json)*

herm0-ct2-int8/config.json

@@ -0,0 +1,10 @@
+{
+  "add_source_bos": false,
+  "add_source_eos": false,
+  "bos_token": "<s>",
+  "decoder_start_token": "</s>",
+  "eos_token": "</s>",
+  "layer_norm_epsilon": null,
+  "multi_query_attention": false,
+  "unk_token": "<unk>"
+}

herm0-ct2-int8/model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d1706ad33f5db7b30bdfc9b6f827426771f9464109ecc13b97ce8ec6f76a79d
+size 57922115

herm0/config.json

@@ -0,0 +1,56 @@
+{
+  "_name_or_path": "/home/user1-gpu/Desktop/grants_folder/windy-pro/models/windy-pair-cus-en/lora",
+  "activation_dropout": 0.0,
+  "activation_function": "swish",
+  "add_bias_logits": false,
+  "add_final_layer_norm": false,
+  "architectures": [
+    "MarianMTModel"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "classif_dropout": 0.0,
+  "classifier_dropout": 0.0,
+  "d_model": 512,
+  "decoder_attention_heads": 8,
+  "decoder_ffn_dim": 2048,
+  "decoder_layerdrop": 0.0,
+  "decoder_layers": 6,
+  "decoder_start_token_id": 23374,
+  "decoder_vocab_size": 23375,
+  "dropout": 0.1,
+  "encoder_attention_heads": 8,
+  "encoder_ffn_dim": 2048,
+  "encoder_layerdrop": 0.0,
+  "encoder_layers": 6,
+  "eos_token_id": 0,
+  "extra_pos_embeddings": 23375,
+  "forced_eos_token_id": 0,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2"
+  },
+  "init_std": 0.02,
+  "is_encoder_decoder": true,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_2": 2
+  },
+  "max_length": null,
+  "max_position_embeddings": 512,
+  "model_type": "marian",
+  "normalize_before": false,
+  "normalize_embedding": false,
+  "num_beams": null,
+  "num_hidden_layers": 6,
+  "pad_token_id": 23374,
+  "scale_embedding": true,
+  "share_encoder_decoder_embeddings": true,
+  "static_position_embeddings": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.45.1",
+  "use_cache": true,
+  "vocab_size": 23375
+}

herm0/generation_config.json

@@ -0,0 +1,16 @@
+{
+  "bad_words_ids": [
+    [
+      23374
+    ]
+  ],
+  "bos_token_id": 0,
+  "decoder_start_token_id": 23374,
+  "eos_token_id": 0,
+  "forced_eos_token_id": 0,
+  "max_length": 512,
+  "num_beams": 4,
+  "pad_token_id": 23374,
+  "renormalize_logits": true,
+  "transformers_version": "4.45.1"
+}

herm0/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00c2cf27f7a9405684f0e6aba1bd78a288f0b5cc5b7efefdaf84f4de2c2b5b78
+size 224548532

herm0/source.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38df7718566f354da8d0b0a91aaf7adb45db623c3de934f324d35c178ed8e22a
+size 453945

herm0/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

herm0/target.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cac993f533fd57d4f0eee80f05406ccae3d463ff1f2968077420d76d295f8b23
+size 450961

herm0/tokenizer_config.json

@@ -0,0 +1,38 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "23374": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "model_max_length": 512,
+  "pad_token": "<pad>",
+  "separate_vocabs": false,
+  "source_lang": "cus",
+  "sp_model_kwargs": {},
+  "target_lang": "eng",
+  "tokenizer_class": "MarianTokenizer",
+  "unk_token": "<unk>"
+}

lora-ct2-int8/config.json

@@ -0,0 +1,10 @@
+{
+  "add_source_bos": false,
+  "add_source_eos": false,
+  "bos_token": "<s>",
+  "decoder_start_token": "</s>",
+  "eos_token": "</s>",
+  "layer_norm_epsilon": null,
+  "multi_query_attention": false,
+  "unk_token": "<unk>"
+}

lora-ct2-int8/model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3deb038c0fcd04152461ccef51d31f1052803f90b05d60bfa52e6d86582852c1
+size 57922115

lora/config.json

@@ -0,0 +1,56 @@
+{
+  "_name_or_path": "Helsinki-NLP/opus-mt-cus-en",
+  "activation_dropout": 0.0,
+  "activation_function": "swish",
+  "add_bias_logits": false,
+  "add_final_layer_norm": false,
+  "architectures": [
+    "MarianMTModel"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 0,
+  "classif_dropout": 0.0,
+  "classifier_dropout": 0.0,
+  "d_model": 512,
+  "decoder_attention_heads": 8,
+  "decoder_ffn_dim": 2048,
+  "decoder_layerdrop": 0.0,
+  "decoder_layers": 6,
+  "decoder_start_token_id": 23374,
+  "decoder_vocab_size": 23375,
+  "dropout": 0.1,
+  "encoder_attention_heads": 8,
+  "encoder_ffn_dim": 2048,
+  "encoder_layerdrop": 0.0,
+  "encoder_layers": 6,
+  "eos_token_id": 0,
+  "extra_pos_embeddings": 23375,
+  "forced_eos_token_id": 0,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2"
+  },
+  "init_std": 0.02,
+  "is_encoder_decoder": true,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_2": 2
+  },
+  "max_length": null,
+  "max_position_embeddings": 512,
+  "model_type": "marian",
+  "normalize_before": false,
+  "normalize_embedding": false,
+  "num_beams": null,
+  "num_hidden_layers": 6,
+  "pad_token_id": 23374,
+  "scale_embedding": true,
+  "share_encoder_decoder_embeddings": true,
+  "static_position_embeddings": true,
+  "torch_dtype": "float32",
+  "transformers_version": "4.45.1",
+  "use_cache": true,
+  "vocab_size": 23375
+}

lora/generation_config.json

@@ -0,0 +1,16 @@
+{
+  "bad_words_ids": [
+    [
+      23374
+    ]
+  ],
+  "bos_token_id": 0,
+  "decoder_start_token_id": 23374,
+  "eos_token_id": 0,
+  "forced_eos_token_id": 0,
+  "max_length": 512,
+  "num_beams": 4,
+  "pad_token_id": 23374,
+  "renormalize_logits": true,
+  "transformers_version": "4.45.1"
+}

lora/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:564509146be62788bf4ab60b66e46fde7864ed4a83b179b8e709c2c2da22a7f8
+size 224548532

lora/source.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38df7718566f354da8d0b0a91aaf7adb45db623c3de934f324d35c178ed8e22a
+size 453945

lora/special_tokens_map.json

@@ -0,0 +1,5 @@
+{
+  "eos_token": "</s>",
+  "pad_token": "<pad>",
+  "unk_token": "<unk>"
+}

lora/target.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cac993f533fd57d4f0eee80f05406ccae3d463ff1f2968077420d76d295f8b23
+size 450961

lora/tokenizer_config.json

@@ -0,0 +1,38 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "23374": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "model_max_length": 512,
+  "pad_token": "<pad>",
+  "separate_vocabs": false,
+  "source_lang": "cus",
+  "sp_model_kwargs": {},
+  "target_lang": "eng",
+  "tokenizer_class": "MarianTokenizer",
+  "unk_token": "<unk>"
+}