When training own dataset, an error occurs when changing numberclasses to the corresponding category. If it is the default, it will report an error
hx358031364 opened this issue · comments
AMP not enabled. Training in float32.
Using native Torch DistributedDataParallel.
Scheduled epochs: 310
/pytorch/aten/src/ATen/native/cuda/ScatterGatherKernel.cu:312: operator(): block: [0,0,0], thread: [15,0,0] Assertion idx_dim >= 0 && idx_dim < index_size && "index out of bounds"
failed.
Traceback (most recent call last):
File "main.py", line 948, in
main()
File "main.py", line 664, in main
optimizers=optimizers)
File "main.py", line 782, in train_one_epoch
output = model(input)
File "/root/miniconda3/lib/python3.7/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "/root/miniconda3/lib/python3.7/site-packages/torch/nn/parallel/distributed.py", line 610, in forward
self._sync_params()
File "/root/miniconda3/lib/python3.7/site-packages/torch/nn/parallel/distributed.py", line 1048, in _sync_params
authoritative_rank,
File "/root/miniconda3/lib/python3.7/site-packages/torch/nn/parallel/distributed.py", line 979, in _distributed_broadcast_coalesced
self.process_group, tensors, buffer_size, authoritative_rank
RuntimeError: CUDA error: device-side assert triggered
terminate called after throwing an instance of 'std::runtime_error'
what(): NCCL error in: /pytorch/torch/lib/c10d/../c10d/NCCLUtils.hpp:136, unhandled cuda error, NCCL version 2.7.8