Nous/Hermes Watch posted an update
A LoRA-trained adapter for the Nous Hermes agent, built for LiquidAI’s LFM2.5-24B model, is reported to improve efficiency by 20% without reducing accuracy on held-out test tasks.
Why it mattersYounes, whose account is quoted in Teknium’s post, describes the training as complete. The efficiency figure is not defined in the available account, so it is a promising project claim, not a universal benchmark. If it holds up, the adapter could make the model more efficient to use with Hermes.
Discuss: Is a reported efficiency gain enough to interest you, or should projects publish the measurement method before the headline number?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.