I launched the vllm server using the DRAM connector with parameter max_cache_size=5368709120. During the first round, I found such warning information:
and obtained the performance results.
However, during the second round, the LLM server crashed and the error messages are: