The six startup lines to read after every start
Six lines from the startup log of vLLM 0.31.0 on a DGX Spark with Qwen3-30B-A3B: version 0.31.0, the V2 model runner, the FLASH_ATTN attention backend, 56.93 GiB of weights loaded in 330 seconds, 31.32 GiB left for the KV cache, and a pool of 234,960 tokens.
LOG LINE
OUR VALUE
WHAT IT DECIDES
version
0.31.0
the release that runs
Using V2 Model Runner
V2
which model runner runs
attention backend
FLASH_ATTN
which kernels run attention
Model loading took
56.93 GiB, 330 s
memory the weights hold
Available KV cache memory
31.32 GiB
memory left for the KV cache
GPU KV cache size
234,960 tokens
how many users fit
Find each line once after every start, before you send traffic.