Dynamic Memory Compression

Originally published at: https://developer.nvidia.com/blog/dynamic-memory-compression/

Despite the success of large language models (LLMs) as general-purpose AI tools, their high demand for computational resources make their deployment challenging in many real-world scenarios. The sizes of the model and conversation state are limited by the available high-bandwidth memory, limiting the number of users that can be served and the maximum conversation length.…