Tag: AI model training
Browse our exclusive articles!
Adaptive Dynamic Sampling for Enhanced Math Reasoning AI
Discover SAI-DPO, a dynamic sampling method that adapts training data to model ability, boosting math reasoning performance efficiently.
Why Output Diversity Collapses in Post-Trained Models
Explore why output diversity collapses in post-trained language models and how training data impacts varied AI responses in creative tasks.
Natural Gradient Descent with Momentum for Faster ML
Discover how combining natural gradient descent with momentum improves convergence and overcomes local minima in machine learning optimization.
StoSignSGD: Optimized SignSGD for Large Language Models
Discover StoSignSGD, an unbiased sign-based optimizer that improves convergence and speeds up training of large language models with better stability.
Subliminal Transfer of Unsafe Behaviors in AI Distillation
Explore how unsafe behaviors subliminally transfer in AI agent distillation, revealing risks beyond explicit data sanitation in model training.