PAR: Progress-Aligned Recurrence for Length Generalization in State Tracking
What this work does
PAR makes recurrent language-model iterations represent successive operations by combining progress-aligned supervision with recurrent-state norm anchoring. Prefix self-distillation supplies intermediate targets when only final answers are available.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.