---
cited_by:
  - '[[thoughts/inference/architectures#3-what-changed-in-vllms-kv-cache-manager-to-hold-these|README §3]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/csrc/libtorch_stable/torch_bindings.cpp
  - file:///Users/aarnphm/workspace/vllm/csrc/libtorch_stable/ops.h
merged: '2026-09-12'
number: 56061
source:
  - https://github.com/vllm-project/vllm/blob/29332cf93628a67c49252ef7c89a8aa3185def4b/vllm/distributed/kv_transfer/kv_connector/v1/hisparse/stats.py
  - https://github.com/vllm-project/vllm/blob/29332cf93628a67c49252ef7c89a8aa3185def4b/vllm/distributed/kv_transfer/kv_connector/v1/hisparse/worker.py
  - https://github.com/vllm-project/vllm/blob/29332cf93628a67c49252ef7c89a8aa3185def4b/vllm/distributed/kv_transfer/kv_connector/v1/hisparse/connector.py
  - https://github.com/vllm-project/vllm/blob/29332cf93628a67c49252ef7c89a8aa3185def4b/csrc/libtorch_stable/hisparse_kernels.cu
  - https://github.com/vllm-project/vllm/blob/29332cf93628a67c49252ef7c89a8aa3185def4b/csrc/libtorch_stable/torch_bindings.cpp
  - https://github.com/vllm-project/vllm/blob/29332cf93628a67c49252ef7c89a8aa3185def4b/vllm/v1/hisparse/runtime.py
  - https://github.com/vllm-project/vllm/blob/29332cf93628a67c49252ef7c89a8aa3185def4b/csrc/libtorch_stable/ops.h
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#56061 [4/N] Expose HiSparse cache metrics via KV connector stats'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/56061
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/56061
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/56061.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Kernel counters (`atomicAdd` in the CUDA kernels) and the swap-ordering changes in `runtime.py` are unchanged from the original PR. · full dump: [[thoughts/inference/vllm-prs/56061|#56061]]

