grimjim/franken-kunoichi-IDUS-11B

Text Generation • Updated Apr 15 • 21

Note Proof by counterexample that fine-tuning is mandatory for coherent behavior after a drastic frankenmerge.

grimjim/kunoichi-squared-model_stock-7B

Text Generation • Updated Apr 15 • 16

Note Unimpressive result of model stock merger. Probably too few models used.

grimjim/wizard-elem-to-32k-7B

Text Generation • Updated May 27 • 6

Note Failed attempt to extend context length to 32k tokens.

grimjim/Llama-3-Perky-Pat-Instruct-8B

Text Generation • Updated Jul 4 • 90 • 4

Note Experiment with negative merge weighting. Basis of proposal for Orthogonal Vector Adaptation (OVA).