Disable CUDA cross-GPU checks when 2 GPUs are used

harvardnlp / seq2seq-attn

Sequence-to-sequence model with LSTM encoder/decoders and attention

http://nlp.seas.harvard.edu/code

MIT License

1.26k stars 278 forks source link

Open macournoyer opened 7 years ago

macournoyer commented 7 years ago

Fixes the following error when training on two GPUs:

Assertion `THCTensor(checkGPU)(state, 3, self, src1, src2)'