chujiezheng commited on
Commit
dd4402b
1 Parent(s): 8416d3f

Add chat_template so that it can be used for chat out-of-box

Browse files

I have added a chat_template in the tokenizer (the new feature of transformers), to facilitate out-of-box use for chat

This template generally follows the claude prompt

Files changed (1) hide show
  1. tokenizer_config.json +1 -0
tokenizer_config.json CHANGED
@@ -9,6 +9,7 @@
9
  "rstrip": false,
10
  "single_word": false
11
  },
 
12
  "clean_up_tokenization_spaces": false,
13
  "eos_token": {
14
  "__type": "AddedToken",
 
9
  "rstrip": false,
10
  "single_word": false
11
  },
12
+ "chat_template": "{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% set system_message = '' %}{% endif %}{% for message in loop_messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if loop.index0 == 0 %}{{ bos_token + system_message }}{% endif %}{% if message['role'] == 'user' %}{{ '\\n\\nUser: ' + message['content'].strip() }}{% elif message['role'] == 'assistant' %}{{ '\\n\\nAssistant: ' + message['content'].strip() }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '\\n\\nAssistant:' }}{% endif %}",
13
  "clean_up_tokenization_spaces": false,
14
  "eos_token": {
15
  "__type": "AddedToken",