In meta_train phase how to change batch_size

[lishuaijun1997]

Thank yours code! I have a Issues, when i use pytorch meta_train phase. I find use batch_sampler will can not use batch_size. My GPU is Nvidia 3070 ,it will CUDA out of memory. Therefore, i hope yaoyao can help me.

[yaoyao-liu]

Thanks for your interest in our work.
May I know what do you mean by "use batch_sampler will can not use batch_size"?
Do you mean that "if you change the batch size, the code is not able to work'?

Best,
Yaoyao

[lishuaijun1997]

self.train_sampler = CategoriesSampler(self.trainset.label, self.args.num_batch, self.args.way, self.args.shot + self.args.train_query) self.train_loader = DataLoader(dataset=self.trainset, batch_sampler=self.train_sampler, num_workers=8, pin_memory=True)
Code only have pre_batch_size,but code have not meta batch_size. So when i use meta_train phase , I can not change meta batch_size.
Thank you for your reply.

[yaoyao-liu]

May I know what do you mean by "meta-batch size"?
If you mean "the number of tasks used to update the meta-model for once", it is one and it cannot be changed in the current PyTorch code.

Best,
Yaoyao

[yaoyao-liu]

The meta-batch size is already 1, so we are not able to reduce it. If you hope to decrease the GPU memory required, you may change the following parameters (i.e., train_query, val_query, and update_step). I don't how large is your GPU memory for your 3070. I recommend you run this project using a GPU with more than 16GB of GPU memory.

[lishuaijun1997]

My GPU memory is 8GB, i will try my lab's TITAN V. I very interest in your work, so I will study your paper carefully!
I truly appreciate your timely help.

[yaoyao-liu]

No problem. If you have any further questions, feel free to send me an email or add comments on this issue.