Question: Does training data order matter at all?

[jsomeara]

I wonder if the order of the data (in https://github.com/tloen/alpaca-lora/blob/main/alpaca_data.json) matters for training.

For example, If I wanted to add my own data to the original dataset in the cleaned JSON file, would it be fine to append it to the end of the JSON? Or should I append and then shuffle the order?

Excuse my ignorance, as I am new to this technology entirely.

Regards

[collant]

I don't think the roder should matter at all, the data is shuffled before training:

From train.py

[jsomeara]

Got it, Thanks!