MiniGPTv2 Train

Finetune of MiniGPT-4

You firstly need to prepare the dataset. you can follow this step to prepare the dataset. our dataset preparation.

In the train_configs/minigptv2_finetune.yaml, you need to set up the following paths:

llama_model checkpoint path: "/path/to/llama_checkpoint"

ckpt: "/path/to/pretrained_checkpoint"

ckpt save path: "/path/to/save_checkpoint"

For ckpt, you may load from our pretrained model checkpoints:

torchrun --nproc-per-node NUM_GPU train.py --cfg-path train_configs/minigptv2_finetune.yaml