Stop token fix?

#1
by americancookie - opened

Hi Michael,

Thanks for your quants, Vielen Dank!

What stop token for LLama 3 quants do you use during the conversion?

Please see the discussion at: https://www.reddit.com/r/LocalLLaMA/comments/1c7dkxh/tutorial_how_to_make_llama3instruct_ggufs_less/

Cheers :)

americancookie changed discussion title from End of token fix? to Stop token fix?

I'm not using any tokens, I merely quantize the model as-is. Also, the "fix" you link to breaks the model, so should not be applied. The problem is in the software, not the gguf. The only correct way to fix this is to update your software for llama-3.

mradermacher changed discussion status to closed

Thank you :)!

Sign up or log in to comment