ResourceExhaustedError (see above for traceback): OOM when allocating tensor of shape [] and type float

[xiaoxiaoAurora]

ResourceExhaustedError (see above for traceback): OOM when allocating tensor of shape [] and type float [[node lm/RNN_0/rnn/multi_rnn_cell/cell_0/lstm_cell/kernel/Initializer/random_uniform/min (defined at /home/miniconda3/lib/python3.6/site-packages/bilm/training.py:410) = Const[_class=["loc:@lm/RNN_0/rnn/multi_rnn_cell/cell_0/lstm_cell/kernel/Assign"], dtype=DT_FLOAT, value=Tensor<type: float shape: [] values: -0.0185652673>, _device="/job:localhost/replica:0/task:0/device:GPU:0"]()]]

How can I solved this?

1. vocabulary size: 146, 828 (including <S>, </S>, <UNK>)
2. I have only one GPU and changed the followings:

• set n_gpu=1,
• export CUDA_VISIBLE_DEVICES=0
  3. NVIDIA-SMI 418.43 Driver Version: 418.43 CUDA Version: 10.1

[matt-peters]

Decrease the batch size or other hyperparameters (vocabulary size, model size, number of negative samples for the softmax) until it fits onto your GPU.