NLP Optimization Comparison Hub
3 papers - avg viability 3.7
Top Papers
- SympFormer: Accelerated attention blocks via Inertial Dynamics on Density Manifolds(5.0)
SympFormer introduces accelerated attention blocks for faster convergence in NLP tasks.
- Gradients Must Earn Their Influence: Unifying SFT with Generalized Entropic Objectives(4.0)
Dynamic Entropy Fine-Tuning (DEFT) offers a novel approach to improve supervised fine-tuning of models by dynamically adjusting token-level weighting according to predictive distribution concentration.
- Efficient Reasoning at Fixed Test-Time Cost via Length-Aware Attention Priors and Gain-Aware Training(2.0)
This research proposes a method for efficient reasoning in Transformers under tight compute constraints.