Update README.md

README.md

@@ -1,19 +1,46 @@
 ---
-library_name: sentence-transformers
-pipeline_tag: sentence-similarity
+pipeline_tag: feature-extraction
 tags:
+- pytorch
 - sentence-transformers
-- feature-extraction
 - sentence-similarity
+- feature-extraction
 - transformers
-
+language:
+- ru
+- en
+datasets:
+- evilfreelancer/opus-php-en-ru-cleaned
+- evilfreelancer/golang-en-ru
+- Helsinki-NLP/opus_books
 ---
 
-# {MODEL_NAME}
+# Enbeddrus v0.2 - English and Russian embedder
+
+> This model trained on Parallel Corpora of Russian and English texts
+
+This is a BERT (uncased) [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 768 dimensional
+dense vector space and can be used for tasks like clustering or semantic search.
+
+- **Parameters**: 168 million
+- **Layers**: 12
+- **Hidden Size**: 768
+- **Attention Heads**: 12
+- **Vocabulary Size**: 119,547
+- **Maximum Sequence Length**: 512 tokens
 
-This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search.
+The Enbeddrus model is designed to extract similar embeddings for comparable English and Russian phrases. It is based on
+the [bert-base-multilingual-uncased](https://huggingface.co/google-bert/bert-base-multilingual-cased) model and was
+trained over 20 epochs on the following datasets:
 
-<!--- Describe your model here -->
+- [evilfreelancer/opus-php-en-ru-cleaned](https://huggingface.co/datasets/evilfreelancer/opus-php-en-ru-cleaned) (train): 1.6k lines
+- [evilfreelancer/golang-en-ru](https://huggingface.co/datasets/evilfreelancer/golang-en-ru) (train): 554 lines
+- [Helsinki-NLP/opus_books](https://huggingface.co/datasets/Helsinki-NLP/opus_books/viewer/en-ru) (en-ru, train): 17.5k lines
+
+The goal of this model is to generate identical or very similar embeddings regardless of whether the text is written in
+English or Russian.
+
+[Enbeddrus GGUF](https://ollama.com/evilfreelancer/enbeddrus) version available via Ollama.
 
 ## Usage (Sentence-Transformers)
 
@@ -27,36 +54,56 @@ Then you can use the model like this:
 
 ```python
 from sentence_transformers import SentenceTransformer
-sentences = ["This is an example sentence", "Each sentence is converted"]
 
-model = SentenceTransformer('{MODEL_NAME}')
+sentences = [
+    "PHP является скриптовым языком программирования, широко используемым для веб-разработки.",
+    "PHP is a scripting language widely used for web development.",
+    "PHP поддерживает множество баз данных, таких как MySQL, PostgreSQL и SQLite.",
+    "PHP supports many databases like MySQL, PostgreSQL, and SQLite.",
+    "Функция echo в PHP используется для вывода текста на экран.",
+    "The echo function in PHP is used to output text to the screen.",
+    "Машинное обучение помогает создавать интеллектуальные системы.",
+    "Machine learning helps to create intelligent systems.",
+]
+
+model = SentenceTransformer('evilfreelancer/enbeddrus-v0.1')
 embeddings = model.encode(sentences)
 print(embeddings)
 ```
 
-
-
 ## Usage (HuggingFace Transformers)
-Without [sentence-transformers](https://www.SBERT.net), you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.
+
+Without [sentence-transformers](https://www.SBERT.net), you can use the model like this: First, you pass your input
+through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word
+embeddings.
 
 ```python
 from transformers import AutoTokenizer, AutoModel
 import torch
 
 
-#Mean Pooling - Take attention mask into account for correct averaging
+# Mean Pooling - Take attention mask into account for correct averaging
 def mean_pooling(model_output, attention_mask):
-    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
+    token_embeddings = model_output[0]  # First element of model_output contains all token embeddings
     input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
     return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
 
 
 # Sentences we want sentence embeddings for
-sentences = ['This is an example sentence', 'Each sentence is converted']
+sentences = [
+    "PHP является скриптовым языком программирования, широко используемым для веб-разработки.",
+    "PHP is a scripting language widely used for web development.",
+    "PHP поддерживает множество баз данных, таких как MySQL, PostgreSQL и SQLite.",
+    "PHP supports many databases like MySQL, PostgreSQL, and SQLite.",
+    "Функция echo в PHP используется для вывода текста на экран.",
+    "The echo function in PHP is used to output text to the screen.",
+    "Машинное обучение помогает создавать интеллектуальные системы.",
+    "Machine learning helps to create intelligent systems.",
+]
 
 # Load model from HuggingFace Hub
-tokenizer = AutoTokenizer.from_pretrained('{MODEL_NAME}')
-model = AutoModel.from_pretrained('{MODEL_NAME}')
+tokenizer = AutoTokenizer.from_pretrained('evilfreelancer/enbeddrus-v0.1')
+model = AutoModel.from_pretrained('evilfreelancer/enbeddrus-v0.1')
 
 # Tokenize sentences
 encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
@@ -72,30 +119,43 @@ print("Sentence embeddings:")
 print(sentence_embeddings)
 ```
 
-
-
 ## Evaluation Results
 
-<!--- Describe how your model was evaluated -->
+The model was tested on the `eval` split of the
+dataset [evilfreelancer/opus-php-en-ru-cleaned](https://huggingface.co/datasets/evilfreelancer/opus-php-en-ru-cleaned),
+which contains 100 pairs of sentences in Russian and English on the topic of PHP. The results of the testing are
+presented in the image below.
 
-For an automated evaluation of this model, see the *Sentence Embeddings Benchmark*: [https://seb.sbert.net](https://seb.sbert.net?model_name={MODEL_NAME})
+![Evaluation Results](./eval.png)
 
+* **Left**: Embedding similarity in Russian and English before training
+  (the points are spread out into two distinct clusters).
+* **Center**: Embedding similarity after training
+  (the points representing similar phrases are very close to each other).
+* **Right**: Cosine distance before and after training.
 
 ## Training
+
 The model was trained with the parameters:
 
 **DataLoader**:
 
-`torch.utils.data.dataloader.DataLoader` of length 573 with parameters:
-```
-{'batch_size': 64, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
+`torch.utils.data.dataloader.DataLoader` of length 556 with parameters:
+
+```python
+{
+    'batch_size': 64,
+    'sampler': 'torch.utils.data.sampler.RandomSampler',
+    'batch_sampler': 'torch.utils.data.sampler.BatchSampler'
+}
 ```
 
 **Loss**:
 
-`sentence_transformers.losses.MSELoss.MSELoss` 
+`sentence_transformers.losses.MSELoss.MSELoss`
 
 Parameters of the fit()-Method:
+
 ```
 {
     "epochs": 20,
@@ -114,8 +174,8 @@ Parameters of the fit()-Method:
 }
 ```
 
-
 ## Full Model Architecture
+
 ```
 SentenceTransformer(
   (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel