So,why train-text-from-scratch and finetune are removed?

[FoxInTango]

I'm trying to merge train-text-from-scratch and finetune from llama.cpp-pretrain to the lastest version of llama.cpp,but errors happened on building.

After checking the codes, I found that the structs -- ggml_tensor, from llama.cpp lastest versioin and llama.cpp-pretrain ,are different.So, I can't continue.

So,why train-text-from-scratch and finetune are removed and how to merge them back to the latest version of llama.cpp?

I really have no idea about that,because I'm not so familiar with the flow of Transformer computing.

[FoxInTango]

So,at last,I found this -- #8669.

Seems a sad story -- expect they will be back soon.

Thanks for all of you.