Update README.md

README.md

@@ -8,4 +8,64 @@ license: apache-2.0
 
 # RuBertTelegramHeadlines
 
-Dataset: https://www.dropbox.com/s/ykqk49a8avlmnaf/ru_all_split.tar.gz
+
+## Model description
+
+Example model for [Headline generation competition](https://competitions.codalab.org/competitions/29905)
+
+## Intended uses & limitations
+
+#### How to use
+
+```python
+
+model_name = "IlyaGusev/rubert_telegram_headlines"
+
+from transformers import AutoTokenizer, EncoderDecoderModel
+
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+hg_model = EncoderDecoderModel.from_pretrained(model_name)
+
+article_text = "..."
+
+input_ids = tokenizer.prepare_seq2seq_batch(
+
+    [article_text],
+    
+    return_tensors="pt",
+    
+    padding="max_length",
+    
+    truncation=True,
+    
+    max_length=256
+    
+)["input_ids"]
+
+output_ids = hg_model.generate(
+
+    input_ids=input_ids,
+    
+    max_length=64,
+    
+    no_repeat_ngram_size=3,
+    
+    num_beams=10,
+    
+    top_p=0.95
+)
+
+headline = tokenizer.decode(output_ids[0], skip_special_tokens=True, clean_up_tokenization_spaces=True)
+
+print(headline)
+
+```
+
+## Training data
+
+- Dataset: [ru_all_split.tar.gz](https://www.dropbox.com/s/ykqk49a8avlmnaf/ru_all_split.tar.gz)
+
+## Training procedure
+
+TBA