sbintuitions

/

sarashina2-8x70b

like 29

Follow

SB Intuitions 55

This PR reduces the confusing about tokenizer loading.

The current setting requires loading the tokenizer with add_eos_token=False or the EOS token will be added automatically, leading to weird completion results.

• Before:

tokenizer = AutoTokenizer.from_pretrained("sbintuitions/sarashina2-8x70b",  add_eos_token=False)

• After:

tokenizer = AutoTokenizer.from_pretrained("sbintuitions/sarashina2-8x70b")

"add_eos_token": false in tokenizer_config.json is the same as sbintuitions/sarashina2-70b's.
https://huggingface.co/sbintuitions/sarashina2-70b/blob/main/tokenizer_config.json#L134