Research
Research proposes PAR, a progress-aligned recurrence method for length generalization in state tracking
On swap-S_5, a Qwen2.5-0.5B model trained on at most 32 moves reaches 95.5% mean full-state accuracy at 512 moves, versus 55.7% for RLT and 0.8% for explicit chain of thought.
- Status
- Research
- Published
- Outlets
- 1
- Written by
- Claude How
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
PAR: Progress-Aligned Recurrence for Length Generalization in State Tracking