UC Irvine

Large language models (LLMs) have emerged and presented their general problem-solvingcapabilities with one model. However, the model size has increased dramatically with billions of parameters to enable such broad problem-solving capabilities. In addition, due to the dominance of matrix-matrix and matrix-vector multiplications in LLMs, the compute-to-model size ratio is significantly lower than that of convolutional neural networks (CNNs). This shift pushes LLMs from a computation-bound regime to a memory-bound regime. Therefore, optimizing the memory footprint and traffic is an important optimization direction for LLMs today.

Model compression methods such as quantization and parameter pruning have been activelyexplored for achieving the memory footprint and traffic optimization. However, the accuracy- efficiency trade-off of rank pruning (i.e., low-rank decomposition) for LLMs is not well- understood yet. Therefore, in this work, we characterize the accuracy-efficiency trade-off of a low-rank decomposition method, Tucker decomposition, on recent language models including an open-source LLM, Llama 2.

We formalize the low-rank decomposition design space and show that the decompositiondesign space is huge (e.g., O(2^37) for Llama2-7B). To navigate such a huge design space, we characterize the design space and prune ineffective design space utilizing the learning from our characterization results (e.g., we can reduce the pruned ranks to 1 without a noticeable model accuracy drop). On the pruned design space, we perform thorough case studies of accuracy-efficiency trade-offs using six widely used LLM benchmarks on BERT and Llama 2 models. Our results show that we can achieve a 9% model size reduction with minimal accuracy drops, which range from 4%p to 10%p, depending on the difficulty of the benchmark, without any retraining to recover accuracy after decomposition. The results show that low-rank decomposition can be a promising direction for LLM-based applications that require real-time service in scale (e.g., AI agent assist and real-time coding assistant), where the latency is as important as the model accuracy.