On the Emergence of Induction Heads for In-Context Learning
"训练动态" 相关文章
The Graphon Limit Hypothesis: Understanding Neural Network Pruning via Infinite Width Analysis
Early-stopping for Transformer model training
Revisiting Meta-Learning with Noisy Labels: Reweighting Dynamics and Theoretical Guarantees
Stability of Transformers under Layer Normalization
What Scales in Cross-Entropy Scaling Law?
Multi-head Transformers Provably Learn Symbolic Multi-step Reasoning via Gradient Descent
Global Convergence and Rich Feature Learning in $L$-Layer Infinite-Width Neural Networks under $\mu$P Parametrization
Improving LLM Reasoning for Vulnerability Detection via Group Relative Policy Optimization
大模型已过时、小模型SLM才是未来？苹果正在研究这个
