Yiming Dong

Conda: Column-Normalized Adam for Training Large Language Models Faster

On the $O(\frac{\sqrt{d}}{K^{1/4}})$ Convergence Rate of AdamW Measured by $\ell_1$ Norm

Stepsize anything: A unified learning rate schedule for budgeted-iteration training

AdaMSS: Adaptive Multi-Subspace Approach for Parameter-Efficient Fine-Tuning

Improving Model Representation and Reducing KV Cache via Skip Connections with First Value Heads