Error in Cross-Entropy Loss Computation with CUDA: Assertion Failure on Batch Size Reduction to 16

1

Hello, I encountered an error while attempting to compute the cross-entropy loss function. The code snippet below presents the relevant portion:

logits_x = logits[:inputs_x.shape[0]]
loss_s = F.cross_entropy(logits_x, targets_x)

Where:

  • targets_x has a size of [16].
  • logits_x has a size of [16, 47].

The specific error message is as follows:

nll_loss_forward_reduce_cuda_kernel_2d: block: [0,0,0], thread: [0,0,0] Assertion `t >= 0 && t < n_classes` failed.
RuntimeError: CUDA error: device-side assert triggered
CUDA kernel errors might be asynchronously reported at some other API call, so the stack trace below might be incorrect. For debugging, consider passing CUDA_LAUNCH_BLOCKING=1.

It’s worth noting that the code runs without any errors on other devices. However, when I reduce the batch size to 16, it results in the aforementioned error. Any insights or suggestions on resolving this issue would be appreciated

2

Your target contains class indices which are out of bounds:

Assertion `t >= 0 && t < n_classes` failed.
3

Thank you. I have reviewed the dataset folder and identified an additional folder that is unrelated to my dataset.