Train it with custom data
Hi there anyone,
I would like to know if there is a way to train zephyr-7b on a local machine with more or less simple data.
For example if there is a website with a store (2k - 10k Links (inc. sublinks) I would like to soak the knowledge up and put it inside this AI to make it to an expert for this Store - is there any way to archive this?
At the moment it feels VERY difficult to do that.
Can someone pls gimme a hug to get a blink of the knowledge I need to get closer to a solution here?
Would be a bless <3
Regards
Manni
Check it out this course.
Regarding train it in a local machine you'll need a good GPU with a minimum of VRAM (just charge the model for inference requires almost 14 GB of VRAM in bfloat16, althugh you can downgrade the precision and train with less memory. This huggingface space is really nice to check the hardware requirements). Perhaps it is a good idea train in google colab or kaggle notebook. I hope it helps!