1 paper · 1 filter
Jing Xiong, Qiujiang Chen, Fanghua Ye +11
Large language models (LLMs) benefit from test-time scaling but are often hampered by high inference latency. Speculative decoding is a natural way to accelerate the scaling proces…