Issues with saving to hub -> Gemma based models

unslothai / unsloth

Finetune Llama 3.2, Mistral, Phi & Gemma LLMs 2-5x faster with 80% less memory

https://unsloth.ai

Apache License 2.0

16.27k stars 1.12k forks source link

Issues with saving to hub -> Gemma based models #1005

Open Ammar-Alnagar opened 3 weeks ago

Ammar-Alnagar commented 3 weeks ago

I have been using unsloth daily for quite sometime now , tried every architecture and there seems to be a problem with gemma based models when saving to hub.

The tokenizer.model file never seems to get saved or pushed to hub no matter what gemma model i use.

Feel free to validate what i am saying through checking my model files and commits and the original model.

My model : Daemontatox/zirel_simp0 Og model : princeton-nlp/gemma-2-9b-it-SimPO

danielhanchen commented 3 weeks ago

Yep known issue - tbh I'm unsure what to do since the latest Gemma models don't seem to pass tokenizer.model properly - I can verify and get back to you! Sorry on the issuse!

Ammar-Alnagar commented 3 weeks ago

No worries , keep up the amazing work