Adapter training reduced embedding size

Adapter training reduced embedding size

Adapter training

MIND trained for 10 epochs on fixed balanced data without user_embedding initialization + contrastive loss

MIND trained for 3 epochs on fixed balanced data without user_embedding initialization + contrastive loss

MIND trained for 3 epochs on fixed balanced data without user_embedding initialization

MIND trained for 2 epochs on balanced data without user_embedding initialization

MIND trained for 90min without user_embedding initialization

MIND initially trained for 30min

One attention head

Initial commit