obs-vllm-block-pool-usage · inference-research

Fields

id

sources/obs-vllm-block-pool-usage

URL

https://github.com/vllm-project/vllm/blob/main/vllm/v1/core/block_pool.py

Publisher

vllm-project/vllm

Kind

repo

Accessed

2026-10-04

notes

Defines exactly what vllm:kv_cache_usage_perc means: get_usage() returns 1.0 - free/total where total = num_gpu_blocks - 1 (the null block is subtracted). It is a block-occupancy ratio computed from the free-block queue, not bytes and not per-request contiguity.

Notes

Defines exactly what vllm:kv_cache_usage_perc means: get_usage() returns 1.0 - free/total where total = num_gpu_blocks - 1 (the null block is subtracted). It is a block-occupancy ratio computed from the free-block queue, not bytes and not per-request contiguity.

Cross-references

This record references no other records.