Clarification on whether breadlicker45/breadchat-save2000 is directly based on ibm-granite/granite-3.3-2b-base

#1
by s1ngledoge - opened

Hi,

Thank you for sharing breadlicker45/breadchat-save2000.

I am considering using this model and wanted to clarify how it relates to ibm-granite/granite-3.3-2b-base before I build on top of it.

Could you please confirm whether breadlicker45/breadchat-save2000 was created directly from ibm-granite/granite-3.3-2b-base through a straightforward fine-tuning step, or whether there were any intermediate checkpoints, additional training stages, merges, distillation steps, or other released models involved in between?

From a practical usage perspective, I am mainly trying to understand whether it should be treated as a direct derivative of ibm-granite/granite-3.3-2b-base, or as a model that went through further modification stages beyond a simple direct fine-tuning path.

This would help me make better compatibility assumptions before using it.

Thank you very much for your time. Any clarification would be greatly appreciated.

Best,
Qu

I did a fully parameter SFT finetune on ibm-granite/granite-3.3-2b-base. My fine-tuning data came from a baking blog site that is mainly focused around bread.

Sign up or log in to comment