2 papers
cs.CL2025
Scaling Laws for Speculative Decoding
Siyuan Yan, Mo Zhu, Guo-qing Jiang +8
The escalating demand for efficient decoding in large language models (LLMs) is particularly critical for reasoning-intensive architectures like OpenAI-o3 and DeepSeek-R1, which de…
cs.CL2025
Multi-Level Optimal Transport for Universal Cross-Tokenizer Knowledge Distillation on Language Models
Xiao Cui, Mo Zhu, Yulei Qin +3
Knowledge distillation (KD) has become a prevalent technique for compressing large language models (LLMs). Existing KD methods are constrained by the need for identical tokenizers…