KL-based self-distillation for large language models
By: Max Rehman Linder
Potential Business Impact:
Teaches computers new words without forgetting old ones.
Large pre-trained language models often struggle to incorporate new domain-specific terminology when fine-tuned on small, specialized corpora. In this work, we address the challenge of vocabulary expansion in frozen LLMs by introducing a mathematically grounded method for knowledge distillation via KL divergence, even when the original and extended models use different tokenizations. This allows the student model to inherit distributional knowledge from the teacher despite differing vocabularies. We compare our KL-based distillation approach to conventional cross-entropy training, evaluating both methods across multiple strategies for initializing new token embeddings. After embedding initialization, models are further fine-tuned to integrate the new vocabulary. Each trained model is benchmarked on approximately 2000 code-generation tasks, where our approach achieves the best performance across the board. Finally, through mechanistic interpretability, we analyze how models learn representations for the new tokens, providing an explanation for the observed gains and offering insight into the structure of embedding space during vocabulary expansion.
Similar Papers
Delta Knowledge Distillation for Large Language Models
Computation and Language
Makes small AI learn better from big AI.
A Dual-Space Framework for General Knowledge Distillation of Large Language Models
Computation and Language
Makes big AI models work in smaller ones.
Knowledge Distillation and Dataset Distillation of Large Language Models: Emerging Trends, Challenges, and Future Directions
Computation and Language
Makes big AI models smaller and faster.