Cuda out of memory
#40
by
comet24082002
- opened
I followed the finetuning example on github to finetune my model for Vietnamese. Then I finetuned it using these parameter but I got error CUDA out of memory. I think that the problem is the token max length is too long ( 8192). I want to ask how to change the token max length to finetune the model to get rid of this problem. Please help me.
You can set query_max_len and passage_max_len to change the max length. For OOM issue, you also can reduce the batch size per_device_train_batch_size.