Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ 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
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

1_Pooling/config.json

@@ -0,0 +1,10 @@
+{
+  "word_embedding_dimension": 1024,
+  "pooling_mode_cls_token": true,
+  "pooling_mode_mean_tokens": false,
+  "pooling_mode_max_tokens": false,
+  "pooling_mode_mean_sqrt_len_tokens": false,
+  "pooling_mode_weightedmean_tokens": false,
+  "pooling_mode_lasttoken": false,
+  "include_prompt": true
+}

README.md

@@ -0,0 +1,161 @@
+---
+language:
+- kk
+- ru
+- en
+license: apache-2.0
+tags:
+- feature-extraction
+- sentence-similarity
+- multilingual
+pipeline_tag: sentence-similarity
+base_model: BAAI/bge-m3
+model-index:
+- name: darmm-embedding-multilingual
+  results:
+    - task:
+        type: retrieval
+        name: Retrieval
+      metrics:
+        - type: recall_at_1
+          value: 0.9444
+        - type: recall_at_3
+          value: 1.0
+        - type: recall_at_5
+          value: 1.0
+        - type: recall_at_10
+          value: 1.0
+---
+
+# Darmm Multilingual Embedding
+
+Multilingual embedding model (Kazakh/Russian/English) fine-tuned from `BAAI/bge-m3` for Darmm FAQ and product content retrieval.
+
+## Usage
+
+Local model usage:
+```python
+from sentence_transformers import SentenceTransformer
+
+model = SentenceTransformer("./outputs/embedding-model")
+sentences = ["Darmm қызметтері қандай?", "What services does Darmm provide?"]
+embeddings = model.encode(sentences)
+print(embeddings.shape)
+```
+
+If published to Hugging Face, replace the path with your repo id.
+
+## Training data (verified)
+- Darmm landing, academy, and mentor site text extracted from local sources.
+Local model usage:
+
+## Training setup
+- Base model: `BAAI/bge-m3`.
+- Loss: `MultipleNegativesRankingLoss` (default in `scripts/train_embeddings.py`).
+- Typical training params in this repo: `epochs=3`, `batch_size=2`, `max_seq_length=128`.
+
+## Evaluation
+Evaluation uses paraphrased FAQ questions mapped to the FAQ corpus:
+- Corpus: `data/faq_chunks.jsonl` (369 chunks)
+- Queries: `data/eval_questions.jsonl` (90 questions)
+
+## Paper & Documentation
+
+<details>
+<summary>🇬🇧 English</summary>
+
+# Darmm: Multilingual Embeddings for FAQ Retrieval
+
+## Abstract
+We present a multilingual embedding model fine‑tuned for Darmm FAQ and product knowledge retrieval in Kazakh, Russian, and English. The model is based on `BAAI/bge-m3` and trained on Darmm website content and a handcrafted FAQ corpus. We evaluate on paraphrased FAQ questions mapped to the FAQ corpus.
+
+## 1. Dataset
+- **Sources**: Darmm landing, academy, and mentor site content (local sources) plus handcrafted FAQ data.
+- **FAQ corpus**: 150 topics × 3 languages = 450 Q/A documents.
+- **Chunked corpus**: 369 chunks in `data/faq_chunks.jsonl`.
+
+## 2. Training
+- **Base model**: `BAAI/bge-m3`
+- **Loss**: `MultipleNegativesRankingLoss`
+- **Params**: `epochs=3`, `batch_size=2`, `max_seq_length=128`
+
+## 3. Results
+Evaluation on `data/eval_questions.jsonl` (90 paraphrased queries) against the FAQ corpus:
+- Recall@1 = 0.9444
+- Recall@3/5/10 = 1.0
+
+## 4. Limitations
+- Performance depends on query style and corpus quality.
+- Short UI strings can reduce relevance; prefer richer FAQ or docs.
+- Validate with real user questions and a held‑out test set.
+
+</details>
+
+<details>
+<summary>🇰🇿 Қазақша</summary>
+
+# Darmm: FAQ іздеуге арналған көптілді эмбеддингтер
+
+## Аңдатпа
+Бұл модель Darmm‑ның FAQ және өнім білім базасын қазақ, орыс және ағылшын тілдерінде іздеуге арналған. Негізі `BAAI/bge-m3`, оқыту Darmm сайт контенті мен қолмен жасалған FAQ жиынына жүргізілді. Бағалау парафраз сұрақтар арқылы жасалды.
+
+## 1. Деректер
+- **Көздер**: Darmm landing/academy/mentor сайттарының локал контенті және FAQ жиыны.
+- **FAQ корпусы**: 150 тақырып × 3 тіл = 450 Q/A құжаты.
+- **Чанкталған корпус**: `data/faq_chunks.jsonl` ішінде 369 чанк.
+
+## 2. Оқыту
+- **Негізгі модель**: `BAAI/bge-m3`
+- **Loss**: `MultipleNegativesRankingLoss`
+- **Параметрлер**: `epochs=3`, `batch_size=2`, `max_seq_length=128`
+
+## 3. Нәтижелер
+`data/eval_questions.jsonl` (90 парафраз сұрақ) арқылы бағалау:
+- Recall@1 = 0.9444
+- Recall@3/5/10 = 1.0
+
+## 4. Шектеулер
+- Нәтиже сұрақ стилі мен корпус сапасына тәуелді.
+- Қысқа UI мәтіндері релевантты төмендетуі мүмкін.
+- Нақты пайдаланушы сұрақтарымен міндетті түрде тексеріңіз.
+
+</details>
+
+<details>
+<summary>🇷🇺 Русский</summary>
+
+# Darmm: Мультиязычные эмбеддинги для FAQ‑поиска
+
+## Аннотация
+Модель предназначена для поиска по FAQ и базе знаний Darmm на казахском, русском и английском. Основана на `BAAI/bge-m3` и дообучена на локальном контенте сайтов Darmm и ручном FAQ‑корпусе. Оценка проводится на перефразированных вопросах.
+
+## 1. Данные
+- **Источ��ики**: локальный контент сайтов Darmm и FAQ‑корпус.
+- **FAQ корпус**: 150 тем × 3 языка = 450 Q/A документов.
+- **Чанкованный корпус**: 369 чанков в `data/faq_chunks.jsonl`.
+
+## 2. Обучение
+- **Базовая модель**: `BAAI/bge-m3`
+- **Loss**: `MultipleNegativesRankingLoss`
+- **Параметры**: `epochs=3`, `batch_size=2`, `max_seq_length=128`
+
+## 3. Результаты
+Оценка на `data/eval_questions.jsonl` (90 перефразированных запросов):
+- Recall@1 = 0.9444
+- Recall@3/5/10 = 1.0
+
+## 4. Ограничения
+- Результаты зависят от стиля запросов и качества корпуса.
+- Короткие UI‑строки снижают релевантность.
+- Проверяйте на реальных пользовательских вопросах.
+
+</details>
+
+## Intended use
+- FAQ search and internal knowledge retrieval across kk/ru/en.
+- RAG pipelines for Darmm services.
+
+## Limitations
+- Results depend on corpus quality and query style.
+- Short UI strings reduce relevance; prefer fuller FAQ or documentation.
+- For real-world validation, use actual user queries and a held‑out test set.

config.json

@@ -0,0 +1,28 @@
+{
+  "_name_or_path": "BAAI/bge-m3",
+  "architectures": [
+    "XLMRobertaModel"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "classifier_dropout": null,
+  "eos_token_id": 2,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "intermediate_size": 4096,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 8194,
+  "model_type": "xlm-roberta",
+  "num_attention_heads": 16,
+  "num_hidden_layers": 24,
+  "output_past": true,
+  "pad_token_id": 1,
+  "position_embedding_type": "absolute",
+  "torch_dtype": "float32",
+  "transformers_version": "4.45.2",
+  "type_vocab_size": 1,
+  "use_cache": true,
+  "vocab_size": 250002
+}

config_sentence_transformers.json

@@ -0,0 +1,9 @@
+{
+  "__version__": {
+    "sentence_transformers": "2.2.2",
+    "transformers": "4.33.0",
+    "pytorch": "2.1.2+cu121"
+  },
+  "prompts": {},
+  "default_prompt_name": null
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59bfcf156b258c3f94a39d494357f8259f38eccf5158d1ebe3a1eecd3a019af1
+size 2271064456

modules.json

@@ -0,0 +1,20 @@
+[
+  {
+    "idx": 0,
+    "name": "0",
+    "path": "",
+    "type": "sentence_transformers.models.Transformer"
+  },
+  {
+    "idx": 1,
+    "name": "1",
+    "path": "1_Pooling",
+    "type": "sentence_transformers.models.Pooling"
+  },
+  {
+    "idx": 2,
+    "name": "2",
+    "path": "2_Normalize",
+    "type": "sentence_transformers.models.Normalize"
+  }
+]

sentence_bert_config.json

@@ -0,0 +1,4 @@
+{
+  "max_seq_length": 128,
+  "do_lower_case": false
+}

sentencepiece.bpe.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cfc8146abe2a0488e9e2a0c56de7952f7c11ab059eca145a0a727afce0db2865
+size 5069051

special_tokens_map.json

@@ -0,0 +1,51 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "cls_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "<mask>",
+    "lstrip": true,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<pad>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c119aa9bc83a5d76efbbc831b23e5790727c12fde474f6519dd96cde6550ffd7
+size 17083052

tokenizer_config.json

@@ -0,0 +1,55 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "3": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "250001": {
+      "content": "<mask>",
+      "lstrip": true,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<s>",
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "<s>",
+  "eos_token": "</s>",
+  "mask_token": "<mask>",
+  "model_max_length": 8192,
+  "pad_token": "<pad>",
+  "sep_token": "</s>",
+  "sp_model_kwargs": {},
+  "tokenizer_class": "XLMRobertaTokenizer",
+  "unk_token": "<unk>"
+}

train_args.json

@@ -0,0 +1,10 @@
+{
+  "pairs": "data/faq_pairs_train.jsonl",
+  "base_model": "BAAI/bge-m3",
+  "output": "outputs/embedding-model",
+  "batch_size": 2,
+  "epochs": 3,
+  "learning_rate": 2e-05,
+  "max_seq_length": 128,
+  "use_hard_negatives": false
+}