I saw some tutorials and finally found the official tutorial is the most easiest to implement <a href="https://pytorch.org/tutorials/beginner/transformer_tutorial.html" rel="noopener noreferrer" target="_blank" style="color: var(--chakra-colors-blue-500); background-color: rgb(255, 255, 255);">https://pytorch.org/tutorials/beginner/transformer_tutorial.html </a> Usually we use blueqat cloud for quantum computing simulation and post API to quantum computers.Now we have cuQuantum and powerful GPUs so I just tried benchmarking some Neural Network tutorials. GPU NVIDIA V100 and CPU Intel(R) CPU @ 2.50GHz 1core Just comparing one epoch benchmarking. GPU<pre class="ql-syntax" spellcheck="false">| end of epoch  1 | time: 72.37s | valid loss 5.80 | valid ppl  331.84
</pre> CPU<pre class="ql-syntax" spellcheck="false">| end of epoch 1 | time: 3313.80s | valid loss 5.82 | valid ppl 336.48
</pre> GPU is about 46 times faster <img src="https://assets.blueqat.com/public/uploads/us-east-2:4805ff4b-c3cc-4344-b165-86544c34d0bf/2023/02/19/Untitled4-ipynb-2-JupyterLab.png"/>

Benchmarking nn.Transformer in blueqat cloud with NVIDIA V100 / CPU

Yuichiro Minato