Missing Chat Template | Finetuning error

[jloadbeach]

Hi together,

I tried to fine-tune the model to my dataset. To do this, I first tried the dummy dataset to get familiar with it. I executed the following command. As a model, I load the 3B model directly from Huggingface. Unfortunately, I get the attached error. If I pass a chat template, I later encounter an error, so I assume that I have passed a wrong template. According to the mapping in the code, the vincuna_v1.5 template must be used, but it does not work. Does anyone have a solution for this?
Thank you!

Command:
bash scripts/NVILA-Lite/sft.sh Efficient-Large-Model/VILA1.5-3b dummy

Error:
...
[rank0]: text = tokenizer.apply_chat_template(
[rank0]: File "/anaconda/envs/vila/lib/python3.10/site-packages/transformers/tokenization_utils_base.py", line 1803, in apply_chat_template
[rank0]: chat_template = self.get_chat_template(chat_template, tools)
[rank0]: File "/anaconda/envs/vila/lib/python3.10/site-packages/transformers/tokenization_utils_base.py", line 1967, in get_chat_template
[rank0]: raise ValueError(
[rank0]: ValueError: Cannot use chat template functions because tokenizer.chat_template is not set and no template argument was passed! For information about writing templates and setting the tokenizer.chat_template attribute, please see the documentation at https://huggingface.co/docs/transformers/main/en/chat_templating

[HoTuanLong]

Hi @jloadbeach, I've met the same issue with you. Have you found a solution?