Assertion srcIndex < srcSelectDimSize failed.

[SparkJiao]

Sorry to bother you
I recently have used your extract_features.py to extract features of some data set but failed. The error information is as follows:
/opt/conda/conda-bld/pytorch_1532584813488/work/aten/src/THC/THCTensorIndex.cu:362: void indexSelectLargeIndex(TensorInfo<T, IndexType>, TensorInfo<T, IndexType>, TensorInfo<long, IndexType>, int, int, IndexType, IndexType, long) [with T = float, IndexType = unsigned int, DstDim = 2, SrcDim = 2, IdxDim = -2, IndexIsMajor = true]: block: [11,0,0], thread: [95,0,0] Assertion srcIndex < srcSelectDimSizefailed. Traceback (most recent call last): File "examples/extract_features.py", line 405, in <module> main() File "examples/extract_features.py", line 375, in main all_encoder_layers, _ = model(input_ids, token_type_ids=None, attention_mask=input_mask) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/torch/nn/modules/module.py", line 477, in __call__ result = self.forward(*input, **kwargs) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/pytorch_pretrained_bert/modeling.py", line 610, in forward output_all_encoded_layers=output_all_encoded_layers) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/torch/nn/modules/module.py", line 477, in __call__ result = self.forward(*input, **kwargs) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/pytorch_pretrained_bert/modeling.py", line 328, in forward hidden_states = layer_module(hidden_states, attention_mask) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/torch/nn/modules/module.py", line 477, in __call__ result = self.forward(*input, **kwargs) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/pytorch_pretrained_bert/modeling.py", line 313, in forward attention_output = self.attention(hidden_states, attention_mask) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/torch/nn/modules/module.py", line 477, in __call__ result = self.forward(*input, **kwargs) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/pytorch_pretrained_bert/modeling.py", line 273, in forward self_output = self.self(input_tensor, attention_mask) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/torch/nn/modules/module.py", line 477, in __call__ result = self.forward(*input, **kwargs) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/pytorch_pretrained_bert/modeling.py", line 224, in forward mixed_query_layer = self.query(hidden_states) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/torch/nn/modules/module.py", line 477, in __call__ result = self.forward(*input, **kwargs) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/torch/nn/modules/linear.py", line 55, in forward return F.linear(input, self.weight, self.bias) File "/home/jiaofangkai/anaconda3/envs/allennlp-env/lib/python3.7/site-packages/torch/nn/functional.py", line 1026, in linear output = input.matmul(weight.t()) RuntimeError: cublas runtime error : resource allocation failed at /opt/conda/conda-bld/pytorch_1532584813488/work/aten/src/THC/THCGeneral.cpp:333
It seems that the index_select function in the models crashed. I read my own data from json files and construct examples from them. I set the batch-size equals 1 and I modified the max_seq_length to the max_length of the input sentences.
Thanks for your help!

[thomwolf]

Your log is very hard to read. Can you format it cleanly?

[SparkJiao]

I'm so sorry
The first error log is as follows:

/opt/conda/conda-bld/pytorch_1532584813488/work/aten/src/THC/THCTensorIndex.cu:362: void indexSelectLargeIndex(TensorInfo<T, IndexType>, TensorInfo<T, IndexType>, TensorInfo<long, IndexType>, int, int, IndexType, IndexType, long) [with T = float, IndexType = unsigned int, DstDim = 2, SrcDim = 2, IdxDim = -2, IndexIsMajor = true]: block: [11,0,0], thread: [95,0,0] Assertion \`srcIndex < srcSelectDimSize\` failed.

And then the Traceback finally points to line 1026 torch/nn/functional.py in linear:
output = input.matmul(weight.t())
It seems that somewhere crashed while using torch.index_select() , do you think it is because my sentence is too long? I will check other aspects, thank you very much

[thomwolf]

It seems like a failed resource allocation.
Maybe you don't have enough RAM or your GPU is too small ?

[SparkJiao]

My GPU has 12400 MB and I think that's enough, may be I should use 'yield' to input the data one by one? I will load less data to try, thanks u a lot!

[thomwolf]

Ok feel free to re-open the issue if you still have troubles.

[zyfedward]

Hi @SparkJiao

I met the same issue here, how did you resolve this?

[nv-quan]

I have the same issue, did you resolve this? @zyfedward @SparkJiao

[LysandreJik]

@nv-quan, do you mind opening a new issue with the template so that we may help?

[SparkJiao]

I have forgot how to reproduce the problem but the index_select error usually happened due to wrong index. You can use a smaller batch size and run the script on CPU to check the full traceback since the traceback while using GPU is delayed.