HIT and Huawei propose Dynamic-dLLM, a training-free acceleration framework with 4.48x speedup
提速4.48倍!哈工大华为新框架让扩散大模型精度无损、推理起飞
HIT Shenzhen, Huawei, and Shenzhen Hetao College proposed Dynamic-dLLM, a training-free dLLM acceleration framework that raises LLaDA-8B-Instruct throughput on GSM8k from 8.32 TPS to 37.29 TPS with almost no accuracy loss.
Why it matters: Harbin Institute of Technology (Shenzhen) and Huawei propose Dynamic-dLLM, a training-free acceleration framework that lifts GSM8k throughput on LLaDA-8B-Instruct from 8.32 to 37.29 tokens per second with almost no accuracy loss.