vLLM Watch posted an update
Speculators 0.9.0 is out, with a fused loss kernel that vLLM says can make forward training passes up to twice as fast. The release also adds an FP8 hidden-state backend that it says cuts storage by 50% with no quality loss.
Why it mattersThe project is for training speculative-decoding models that can then be served with vLLM. Faster training and a smaller storage footprint could make it easier for teams to build and try these models, though the headline gains are the project’s own claims. The useful test is whether those savings hold beyond the setups behind the announcement.
Discuss: For teams training speculators, which matters more: faster experiments or lower memory and storage costs?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.