Axolotl Configs Settings

#9
by GokhanAI - opened

Hello, First of all, I would like to thank you for your work. If possible, would you be able to share the Axolotl Configs settings? I wish you good work.

Hi @teknium ,
Following up on this.
Can you please share the Axolotl Configs settings? Thanks!

at least the special tokens part :)

NousResearch org

at least the special tokens part :)

special_tokens:
pad_token: <|end_of_text|>
eos_token: <|im_end|>

but no <|im_start|> ?

@cinjonr i would say yes. I personally trained another llama3 with the following:

special_tokens:
eos_token: "<|im_end|>"
pad_token: "<|end_of_text|>"
tokens:

  • "<|im_start|>"

if you pay for training cloud GPU services I strongly suggest to make a test first with a few rows, test inference to be sure the model does stop the generation.

NousResearch org

but no <|im_start|> ?

No, no imstart token as a special token. It is an added token though, which you can use this as a reference for:
https://huggingface.co/NousResearch/Meta-Llama-3-8B-Alternate-Tokenizer

teknium changed discussion status to closed

Sign up or log in to comment