MoDification: Mixture of Depths Made Easy
This work addresses efficiency challenges in serving LLMs for long-context tasks, offering a practical solution with incremental improvements over prior MoD methods.
The paper tackled the problem of efficiently transforming existing large language models (LLMs) into mixture of depths (MoD) variants without costly training, achieving up to ~1.2x speedup in latency and ~1.8x reduction in memory for long-context applications.
Long-context efficiency has recently become a trending topic in serving large language models (LLMs). And mixture of depths (MoD) is proposed as a perfect fit to bring down both latency and memory. In this paper, however, we discover that MoD can barely transform existing LLMs without costly training over an extensive number of tokens. To enable the transformations from any LLMs to MoD ones, we showcase top-k operator in MoD should be promoted to threshold-p operator, and refinement to architecture and data should also be crafted along. All these designs form our method termed MoDification. Through a comprehensive set of experiments covering model scales from 3B to 70B, we exhibit MoDification strikes an excellent balance between efficiency and effectiveness. MoDification can achieve up to ~1.2x speedup in latency and ~1.8x reduction in memory compared to original LLMs especially in long-context applications.