trouble_shooting.md

일반적인 상황

RuntimeError: a leaf Variable that requires grad is being used in an in-place operation.

for name, param in model.named_parameters():
param += 3

에서 for name, param in model.named_parameters():
param.data += 3

로 수정한다

RuntimeError: CUDA error: device-side assert triggered

device cpu로 바꾸고 디버깅 진행하기

IndexError: Target 3226 is out of bounds.

target의 min, max 를 확인해본다

TypeError: default_collate: batch must contain tensors, numpy arrays, numbers, dicts or lists; found <class 'torchtext.data.example.Example'>

TabularDataset은 torchtext의 Iterator와 연동되고, DataLoader랑 하려면 DataSet을 써야함

expected scalar type long but found float

target = target.long()

The target should be a LongTensor using nn.CrossEntropyLoss (or nn.NLLLoss), since it is used to index the output logit (or log probability) for the current target class as shown in this formula 422 (note the indexing in x[class]).

new() received an invalid combination of arguments - got (int, dtype=type), but expected one of: * (*, torch.device device) didn't match because some of the keywords were incorrect: dtype * (torch.Storage storage) * (Tensor other) * (tuple of ints size, *, torch.device device) * (object data, *, torch.device device)

dtype을 int로 주고싶은 경우 torch.Tensor(xx, dtype=torch.int64) 가 아니라 torch.Tensor(xx).long()와 같이 짜기

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation

requires_grad=true주기?

Thre were no tensor arguments to this function (e.g., you passed an empty list of Tensors), but no fallback function is registered for schema aten::_cat. This usually means that this function requires a non-empty list of Tensors. Available functions are [CPU, CUDA, QuantizedCPU, BackendSelect, Named, AutogradOther, AutogradCPU, AutogradCUDA, AutogradXLA, AutogradPrivateUse1, AutogradPrivateUse2, AutogradPrivateUse3, Tracer, Autocast, Batched, VmapMode].

빈 리스트에 대해서 텐서 연산을 했을 경우.

only one element tensors can be converted to python scalars

dim이 (n,)인 tensor가 있는 list를 torch.Tensor혹은 np.array로 만들때, 에러가 남 dim을 .unsqueeze로 (n, 1)로 만들어 주니 해결됨

RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.

AttributeError: cannot assign module before Module.init() call

super().init()를 추가하면 된다

데이터셋을 만들었는데 데이터 로더에 넣었는데 인덱스 에러가 나요

두 페어의 길이가 다른 것임. 데이터셋 잘못 만들었을 것

pytorch/aten/src/ATen/native/cuda/Indexing.cu:658: indexSelectLargeIndex: block: [67,0,0], thread: [0,0,0] Assertion srcIndex < srcSelectDimSize failed.

임베딩 input차원이 실제 들어간 것보다 작음. input차원 맞게 했는지 확인하기.

mlflow: RuntimeError: Cannot set the device explicitly. Please use module.to(new_device).

pl.LightningModule을 상속한 경우 self.device를 주면 안되는 듯 하다. 다른 이름으로 넣어주자