Why Is Speculative Decoding Fast?

50 points by gmays a day ago on hackernews | 5 comments

we have action chunking in robotics for the same reason so VLAs can analyze faster and come to actions, the latency and inference are both slow so many steps need to come together for this to happen correctly

ArashEdalat | 9 hours ago

has anyone tried different approaches to this?

: guipsp | 9 hours ago
...yes? Can you put more effort into your question?
: RugnirViking | 8 hours ago
this morning I tried a different approach to speculative decoding by not writing any code at all and in fact going out for a walk around the lake near my house? What do you mean specifically?

jbarrow | 5 hours ago

This is my post! Happy to answer any follow-up questions people have.

I wrote this to try and clear up what I think is a common misconception (speculative decoding is a free lunch), when the picture is more nuanced (and imo, interesting).