OOM on 24gb GPU (4090) when running training tutorial

```
RuntimeError: CUDA out of memory. Tried to allocate 4.00 GiB (GPU 0; 23.64 GiB total capacity; 15.74 GiB already allocated; 1.41 GiB free; 19.97 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
```

Trying to run the training tutorial out of the box - how much VRAM is needed?

Start by reviewing the training tutorial referenced in the issue and reproduce it with the reported 24 GB GPU if possible. Record the VRAM requirement or limitations exposed by the CUDA out-of-memory error; done means the tutorial documents its hardware requirements clearly.

Written by the indexing model from the issue text.