Add chat template

README.md

@@ -48,6 +48,30 @@ Your prompt here
 The output of Stable Beluga 7B
 ```
 
+This formatting is also available via a pre-defined Transformers chat template, which means that lists of messages can be formatted for you with the `apply_chat_template()` method:
+
+```python
+chat = [
+  {"role": "system", "content": "This is a system prompt, please behave and help the user."},
+  {"role": "user", "content": "Your prompt here"},
+]
+tokenizer.apply_chat_template(chat, tokenize=False)
+```
+
+which will yield:
+
+```
+### System:
+This is a system prompt, please behave and help the user.
+
+### User:
+Your prompt here
+
+
+```
+
+If you use `tokenize=True` and `return_tensors="pt"` instead, then you will get a tokenized and formatted conversation ready to pass to `model.generate()`.
+
 ## Model Details
 
 * **Developed by**: [Stability AI](https://stability.ai/)
@@ -96,4 +120,4 @@ Beluga is a new technology that carries risks with use. Testing conducted to dat
       archivePrefix={arXiv},
       primaryClass={cs.CL}
 }
-```
+```

tokenizer_config.json

@@ -7,6 +7,7 @@
     "rstrip": false,
     "single_word": false
   },
+  "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{ '### ' + message['role'].title() + ':\n' + message['content'] + '\n\n' }}{% endfor %}{% if add_generation_prompt %}{{ '###Assistant:\n' }}{% endif %}",
   "clean_up_tokenization_spaces": false,
   "eos_token": {
     "__type": "AddedToken",