News

Prior versions of the image captioning model took three seconds per training step on an Nvidia G20 GPU, but the version open sourced today can do the same task in a quarter of that time, or just 0 ...