Loading the Newsroom

The open source engine is tightening contributor rules to keep “slop” from drowning maintainers and killing mentorship.

A new MIT-licensed speculative decoding framework raises throughput and per-user token speed without changing the target model.