Error training yolov8 for segmentation on a new custom model Assertion `index >= -sizes[i] && index < sizes[i] && "index out of bounds"

Question

I am training yolov8 for segmnentation with this command:

!yolo task=segment mode=train model=yolov8m-seg.pt data={HOME}/new_data/data.yaml epochs=500 imgsz=640 batch=8 workers=2

and this is how the label txt files look like: 32 0.03567479464644724 0.03869047619047619 0.04130750547359965 0.9375 1.0 0.9315476190476188 0.9857524195589085 0.004464285714285714

We have 33 classes of objects.

But this error raises:

ultralytics/utils/tal.py", line 152, in get_box_metrics
    bbox_scores[mask_gt] = pd_scores[ind[0], :, ind[1]][mask_gt]  # b, max_num_obj, h*w
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.
Compile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.

I checked the annotations but still no answer for this error!

These are the efforts that I made:

1. Checking the size of the dataset

2. Finding files with invalid bounding boxes

3. Checking missing files

4. Verifying the images

5. Verifying the labels

6. Verifying the 33 classes

7. Checking the labels for any inconsistency such as being out of bounds and not being normalised, negative labels, duplicate labels, ...

The results didn't show any reason why this is happening.