Curiosity-Driven Quantized Mixture-of-Experts for Stable AI
Discover a curiosity-driven quantized Mixture-of-Experts framework that boosts AI stability, accuracy, and energy efficiency on resource-limited devices.
QUARK: Efficient Transformer Acceleration via Quantization
Discover QUARK, a quantization-enabled framework that accelerates transformers by sharing circuits in nonlinear operations, boosting speed and reducing har...
Robust Multilingual LLM Watermarking via Back-Translation
Discover how STEAM enhances multilingual LLM watermarking robustness across 100+ languages using back-translation and Bayesian optimization.
OffSim: Advanced Offline Simulator for Inverse RL
Discover OffSim, an offline simulator enhancing model-based inverse reinforcement learning with expert data for efficient policy training.
Future Summary Prediction: Advancing LLM Pretraining
Discover how future summary prediction enhances large language models by improving long-term reasoning and creative text generation beyond token prediction...