princeton-nlp
commited on
Commit
•
fe8c3ca
1
Parent(s):
364a199
Update tokenizer_config.json
Browse filesThe chat template has been corrected. The model is trained and evaluated using the Zephyr (https://huggingface.co/HuggingFaceH4/zephyr-7b-beta/blob/main/tokenizer_config.json) chat template. However, we previously mistakenly uploaded the Mistral-Instruct (https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/blob/main/tokenizer_config.json) template.
- tokenizer_config.json +1 -1
tokenizer_config.json
CHANGED
@@ -29,7 +29,7 @@
|
|
29 |
},
|
30 |
"additional_special_tokens": [],
|
31 |
"bos_token": "<s>",
|
32 |
-
"chat_template": "{
|
33 |
"clean_up_tokenization_spaces": false,
|
34 |
"eos_token": "</s>",
|
35 |
"legacy": true,
|
|
|
29 |
},
|
30 |
"additional_special_tokens": [],
|
31 |
"bos_token": "<s>",
|
32 |
+
"chat_template": "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n' + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}",
|
33 |
"clean_up_tokenization_spaces": false,
|
34 |
"eos_token": "</s>",
|
35 |
"legacy": true,
|