I am trying to train a yolov11 model for a set of images but I am running into an issue. The GPU the model is training on is a GeForce RTX 4070 super.

I am using yolov11 with torch with GPU.

print(torch.__version__)
2.5.1+cu121

I did train the model a few times before without running into issues with this command:

yolo task=detect mode=train device=0 epochs=2000 batch=32 data="C:\Project\Yolo\data_custom.yaml" model="C:\Project\Yolo\yolov11m.pt" imgsz=640

But this did not give me results that are good enough. So if I understand proprely, by increasing the value of "imgsz" to 1440 (my images are 2560 by 1440), the model should train with images of grater quality. So I tried to run this command:

yolo task=detect mode=train device=0 epochs=2000 batch=32 data="C:\Project\Yolo\data_custom.yaml" model="C:\Project\Yolo\yolov11m.pt" imgsz=640

Running this command gives me an error that I don't know how to solve. The error is the following:

torch.OutOfMemoryError: CUDA out of memory. Tried to allocate 1.48 GiB. GPU 0 has a total capacity of 11.99 GiB of which 0 bytes is free. Of the allocated memory 19.12 GiB is allocated by PyTorch, and 406.58 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True to avoid fragmentation.  See documentation for Memory Management  (.html#environment-variables)

Does anyone how to solve this issue ?