Train/Validation splits

[zhangliang-04]

Dear authors, Thank you for the great work and open sourcing. I noticed that you keep about 1k validations for alpaca, llava, quora, and sharegpt #39. Since I want to keep the same setting in my experiments, would you mind sharing the splits with me?

[MAGAer13]

We randomly shuffle the data with random seed 42 and split into the training and validation split.