numerical stability
Numerical stability refers to the property of algorithms to produce accurate results despite small perturbations in the input data or intermediate calculations. In AI, this is crucial for ensuring that optimization algorithms converge correctly without yielding erratic solutions.
- Constrained Entropic Unlearning: A Primal-Dual Framework for Large Language Models
- Efficient Parametric SVD of Koopman Operator for Stochastic Dynamical Systems
- LoRASuite: Efficient LoRA Adaptation Across Large Language Model Upgrades
- Metric Automata Theory: A Unifying Theory of RNNs
- Understanding and Mitigating Numerical Sources of Nondeterminism in LLM Inference
- Understanding and Mitigating Numerical Sources of Nondeterminism in LLM Inference