How many data exactly are used in stage-2 fintuning in MiniGPT-4 v1? I'm confused.

[Oklahomawhore]

In train_configs/minigpt4_llama2_stage2_finetune.yaml :

  iters_per_epoch: 200
  batch_size_train: 12
  accum_grad_iters: 6

Since the code uses a MultiIterDataloader which turns the data loader into a infinite loop. Because my 4090 can't fit MiniGPT4 stage2 finetune, I'm changing the batch size to 2.
Does that mean I'm using 200x2x5=2000 data points only? How do I change accum_grad_iters accordingly?

from runner/runner_base.py:

for cur_epoch in range(self.start_epoch, self.max_epoch):
            # training phase
            if not self.evaluate_only:
                logging.info("Start training")
               

from tasks/base_task.py:

for i in metric_logger.log_every(range(iters_per_epoch), log_freq, header):
            # if using iter-based runner, we stop after iters_per_epoch iterations.
            if i >= iters_per_epoch:
                break

            samples = next(data_loader)

[waltonfuture]

Same questions @junchen14