[wills_forward]

MIT and Microsoft Researchers Introduce "RetNet" - An 8X Faster Transformer Alternative for AI

[mirekrusin]

The claim is parallelism for training which is not fixed speed up, different complexity for inference (constant time), and different complexity for large context inference (linear) - so nothing that can be summarised as 8x - or am I getting this summary wrong?

[whimsicalism]

the words per second i believe from the first graph in the paper

[canjobear]

What’s the MIT connection? The authors all seem to be affiliated with Tsinghua.