d26 models with the pirate-2x2 corpora inserted in pretraining: exp-056 anchor + exp-074 dose/window sweep, base and instruction-SFT pairs.
Julian Minder
jkminder
AI & ML interests
None yet
Recent Activity
published a model about 1 hour ago
jkminder/d12_110m_seed1 updated a model about 4 hours ago
jkminder/d12_110m_seed1 updated a collection 4 days ago
Pretraining priors: pirate-2x2 planted prior (d26)