Benchmarks & research
TurinTech compresses Clef, Decider and Kev into Artemis files for llama.cpp, trading memory use against how closely their answers match the original models.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
These independent compressed derivatives are not Jev weights or endorsements from the original makers. TurinTech reports how often each file retains its source model's answers; agreement also retains the source's mistakes. Its evaluation informed file selection, so it does not establish performance on unseen inputs.
Choose a file that fits your available memory and run its documented decision service, not a chat-model button. The Clef instructions require both batch settings to be at least 4,096. Smaller settings can reject a long request even when the model itself has room for it.