Provide high-performance storage for LLM training and KV offload to improve training efficiency.
Teams may use general-purpose distributed storage or local SSDs, but performance is insufficient.
Existing storage systems may not meet the throughput and latency requirements of LLM training, leading to inefficiency.