CUDA out of memory issue

[eladrich]

CUDA out of memory issue

[flow-specter]

I have the same issue...

[qsong2001]

if you met this issue:
RuntimeError: CUDA error: an illegal memory access was encountered

You can simply try to re-run the training script repeatedly (like 4,5 times) . It worked for me.

[SzaremehrjardiMT]

Same issue on 4GB RAM.
torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 26.00 MiB (GPU 0; 4.00 GiB total capacity; 3.38 GiB already allocated; 0 bytes free; 3.42 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
I tried different solutions such as reducing batch size, using torch.no_grad() for Inference and loading models in float16 precision instead of the default float32 precision; But none of them worked for me.

What is the minimum hardware requirements?