Too BIG! max_model_length in tokenizer_config.json
#1
by
Hambaobao
- opened
max_model_length in tokenizer_config.json seems wrong. Why is it so big, I see max_model_length in other reps is only 2048.
@Hambaobao
yeah I had this issue too when I wrote a script to pad my datasets in max_model_length, boy that was insane to have 1000000000000000019884624838656 length. I made a pull request in https://huggingface.co/NousResearch/Llama-2-7b-hf/discussions/2/files to make it 4k
It doesn't actually matter so long as its 4096 or greater, but it is confusing. Updating
teknium
changed discussion status to
closed