InstCache: A Predictive Cache for LLM Serving

Large language models are revolutionizing every aspect of human life. However, the unprecedented power comes at the cost of significant computing intensity, suggesting long latency and large energy footprint. Key-Value Cache and Semantic Cache have been proposed as a solution to the above problem, but both suffer from limited scalability due to significant memory cost for each token or instruction embeddings. Motivated by the observations that most instructions are short, repetitive and predictable by LLMs, we propose to predict user-instructions by an instruction-aligned LLM and store them in a predictive cache, so-called InstCache. We introduce an instruction pre-population algorithm based on the negative log likelihood of instructions, determining the cache size with regard to the hit rate. The proposed InstCache is efficiently implemented as a hash table with minimal lookup latency for deployment. Experimental results show that InstCache can achieve up to 51.34% hit rate on LMSys dataset, which corresponds to a 2x speedup, at a memory cost of only 4.5GB.

CQIL: Inference Latency Optimization with Concurrent Computation of Quasi-Independent Layers

The fast-growing large scale language models are delivering unprecedented performance on almost all natural language processing tasks. However, the effectiveness of large language models are reliant on an exponentially increasing number of parameters. The overwhelming computation complexity incurs a high inference latency that negatively affects user experience. Existing methods to improve inference efficiency, such as tensor parallelism and quantization, target to reduce per-layer computing latency, yet overlook the cumulative latency due to the number of layers. Recent works on reducing the cumulative latency through layer removing, however, lead to significant performance drop. Motivated by the similarity of inputs among adjacent layers, we propose to identify quasi-independent layers, which can be concurrently computed to significantly decrease inference latency. We also introduce a bypassing technique to mitigate the effect of information loss. Empirical experiments of the proposed approach on the LLaMA models confirm that Concurrent Computation of Quasi-Independent Layers (CQIL) can reduce latency by up to 48.3% on the LLaMA-33B model, while maintaining a close level of performance.