/opt/conda/conda-bld/pytorch_1708025847130/work/aten/src/ATen/native/cuda/Loss.cu:250: nll_loss_forward_reduce_cuda_kernel_2d: block: [0,0,0], thread: [14,0,0] Assertion `t >= 0 && t < n_classes` failed.

[yaohwang]

when I tried to train some 'qna' style dataset like knowrohit07/know_sql get this error.