3 ms·
training took over 7 hours, uses only 1 core, scope for improvement பிற்பகை பொருள்தெரியா மன்னவன் கொற்றங் கொளக்கிடந்தது இல்.
by senthilnayagam 4y ago
training took over 7 hours, uses only 1 core, scope for improvement
பிற்பகை பொருள்தெரியா மன்னவன்
கொற்றங் கொளக்கிடந்தது இல்.
- codetiger 4y agoWhat I noticed was, block_size config 16 works best. Increasing it gives mostly the original kural. There are lot of ideas for further optimisation, especially in the encoder/decoder as it is not tuned for Tamil.
- codetiger 4y agoI tried configuring the training for Apple M1 chip and I was able to configure the training to work on both CPU / GPU based on how you configure. In my case it was faster on CPU. If you are running on other platforms, We need to check on PyTorch config for utilising better CPU.