Sakana AICycleQD

Llama-3-8B-Instruct-CycleQD-CS

CycleQD-CS combines Sakana's three task experts through quality-diversity evolution.

Research release

Evolution combines skills in one checkpoint

CycleQD uses model merging and singular-value mutation, cycling which skill defines quality.Sakana AI CycleQD training method

Evaluate all three task types

Practical guidance. Compare the merged model with each expert on its own specialty. An average score can hide a loss in database work even when coding improves.

Specifications & access

Llama-3-8B-Instruct-CycleQD-CS specifications
DetailValue and conditions
StatusResearch release
AccessDownloadable
Model identifiersSakanaAI/Llama-3-8B-Instruct-CycleQD-CS
AncestryLlama 3 8B Instruct through the OS, DB and Coding experts.Llama-3-8B-Instruct-CycleQD-CS official model card
TermsLlama 3 license; research prototype.Llama-3-8B-Instruct-CycleQD-CS official model card

Related models