Llama adds this extra token when the first character is '\n', and this compromises the stopping criteria, so we just remove it

[ArEnSc]

Describe the bug

Was looking at the code auditing it and I came across this.
But this does nothing since 29871 is the letter \ and 13 is n and character and it adds a '\n' to the string after.
So it makes me wonder if it works fine without this code anyways?
the token it adds is 1 or btw.

Llama adds this extra token when the first character is '\n', and this

compromises the stopping criteria, so we just remove it

if type(shared.tokenizer) is transformers.LlamaTokenizer and input_ids[0][0] == 29871:
input_ids = input_ids[:, 1:]

Is there an existing issue for this?

• I have searched the existing issues

Reproduction

The code should be
if type(shared.tokenizer) is transformers.LlamaTokenizer and input_ids[0][1] == 29871:
input_ids = input_ids[:, 1:]

to remove the character.

Screenshot

No response

Logs

none

System Info

none

[ArEnSc]

but wouldn't we want that token to start or be a terminating condition ?