Training 3x larger model on the same GPU cards

Y	Hacker News new \| ask \| show \| jobs

	Training 3x larger model on the same GPU cards (github.com)
	2 points by xxr3376 1848 days ago

1 comments

MegEngine (A Deep Learning Framework) implements DTR. Now you can train 3x larger model by tradeoff a little bit speed for lots of GPU memory.