环境： K100AI vllm 0.8.5.post1+das.opt1.dtk25041 torch 2.4.1+das.opt1.dtk25041 非docker，运行命令 export VLLM_USE_V1=1 # 强制开启v1引擎，效果更好，但是还在完善中 export VLLM_USE_TRITON_AWQ=1 # 设置模型的量化方式 export HIP_VISIBLE_DEVICES=0 # 只使用 GPU 0 和 1 export VLLM_WORKER_MULTIPROC_METHOD=spawn export VLLM_MARLIN_USE_ATOMIC_ADD=1 vllm serve /public/home/scncrbig3v/qwen_awq --disable-log-requests --host 0.0.0.0 --port 8080 --served-model-name qwen2.5 --max-num-seqs 8 --block-size 32 --gpu-memory-utilization 0.8 --max-model-len 8192 --tensor-parallel-size 1 --chat-template-content-format auto --enable-chunked-prefill --dtype float16 INFO: 172.20.193.237:60610 - "POST /v1/chat/completions HTTP/1.1" 200 OK INFO 10-15 17:00:29 [loggers.py:111] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 1.4 tokens/s, Running: 0 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.3%, Prefix cache hit rate: 0.0% INFO 10-15 17:00:39 [loggers.py:111] Engine 000: Avg prompt throughput: 0.0 tokens/s, Avg generation throughput: 0.0 tokens/s, Running: 0 reqs, Waiting: 0 reqs, GPU KV cache usage: 0.3%, Prefix cache hit rate: 0.0% INFO 10-15 17:01:29 [loggers.py:111] Engine 000: Avg prompt throughput: 27.9 tokens/s, Avg generation throughput: 1.8 tokens/s, Running: 1 reqs, Waiting: 0 reqs, GPU KV cache usage: 3.2%, Prefix cache hit rate: 50.0% INFO: 172.20.193.237:37698 - "POST /v1/chat/completions HTTP/1.1" 200 OK 相同的模型，相同图片，相同的text 4090 要2.5s k100ai需要8s左右，差距有点大。 是我那个地方没有配置对吗？