I have downloaded pre-trained models BERT, llama-2 and some others (for CPU). I wanted to test, on the h2ogpt platform what output would each provide.
My problem is how the knowledge of the models work after the training.
What exactly to do with the datasets?
Example the model is pre-trained on massive data and works as in gpt.h2o.ai.
If I change to personal MyData Collection:
If there are not any i become not any answers. If I use some data I become some answers, with lot errors. If I use lot of data i become no answers. If I use LLM without userdata I become perfect answers