Tag: LLM training
Browse our exclusive articles!
SPICE: Efficient Data Selection for Large Language Model Training
SPICE improves large language model training by selecting conflict-aware data subsets, boosting performance while reducing training costs significantly.
Weakly Supervised Hallucination Detection in Transformers
Discover a novel weak supervision method to distill hallucination signals into transformer models for efficient, internal hallucination detection without e...
TRACE: Advanced Capability-Targeted Training for LLMs
Discover TRACE, a system enhancing LLMs with targeted training environments for improved agent capabilities and efficiency in diverse tasks.
Boost LLM Learning with Vocabulary Dropout for Diversity
Improve language model training with vocabulary dropout to prevent diversity collapse and enhance autonomous curriculum learning in LLM co-evolution.
Optimsyn: Optimizing Synthetic Data with Influence-Guided Rubrics
Discover Optimsyn, a framework that enhances synthetic data generation using influence-guided rubric optimization and reinforcement learning for better mod...