All pull requests

Pull requests

• feat(radixshmem): add asynchronous peer push through FlexKV

  #313·gxr1022 opened Oct 12, 2026·

  ·

• feat(kvevent): report host-tier placement to KV-cache-aware routers

  #310·linhu-nv opened Oct 10, 2026Collaborator·

  ·

• Feature: Add CXL Type 3 Memory Tier Support

  #308·thsrini opened Oct 7, 2026·

  ·
  1

• build(packaging): make the source distribution installable from PyPI

  #301·linhu-nv opened Sep 23, 2026Collaborator·

  ·

• nvcomp: compress multiple KV layer groups, selected by dtype headroom

  #298·linhu-nv opened Sep 21, 2026Collaborator·

  ·

• perf(bindings): release GIL for native transfers

  #284·hu-hamster opened Sep 2, 2026Contributor·

  ·
  6

• [feature] add StreamingLLM sink-aware eviction

  #271·Claisenn opened Aug 24, 2026Contributor·

  ·
  2

• feat: per-node PP pool offset with per-stage and multi-group transfer correctness

  #256·zhjc1124 opened Aug 10, 2026Contributor·

  ·

• fix(sglang): auto-derive dp_rank for dp_attention mode

  #243·linhu-nv opened Jul 29, 2026Collaborator·

  ·

• support merge to batch when using gds

  #239·linhu-nv opened Jul 28, 2026Collaborator·

  ·

• Dsv4 metrics on main

  #234·LfieLike opened Jul 26, 2026·

  ·

• fix(mooncake): defer cache publication for remote loads and concurrent puts

  #232·XingLiu1 opened Jul 23, 2026Collaborator·

  ·

• Sglang + flexkv to run glm5.2 with nvfp4 kvcache type

  #226·linhu-nv opened Jul 16, 2026Collaborator·

  ·

• Moritzxu/swa refactor

  #224·Luis-xu opened Jul 15, 2026Collaborator·

  ·
  1

• Add Mooncake Store backend support and SGLang compatibility aliases

  #217·LfieLike opened Jul 13, 2026·

  ·

• Add and harden FlexKV stress benchmark

  #214·XingLiu1 opened Jul 12, 2026Collaborator·

  ·

• use single concrete FlexKVConnectorStats subclass for worker and scheduler

  #189·linhu-nv opened Jun 26, 2026Collaborator·

  ·

• feat(dist_reuse): KV cache sharing across TP/PP/CP + single-node radi…

  #169·feiqiangs opened May 19, 2026Collaborator·

  ·
  14

• Prefetch optimize, adding three prefetch strategies

  #163·Luis-xu opened May 12, 2026Collaborator·

  ·

• use wider copy granuality and bypass L1

  #157·linhu-nv opened Apr 29, 2026Collaborator·

  ·

• [feature] Add SGLang HiCacheStorage integration with local and distributed modes

  #137·staryxchen opened Apr 5, 2026Collaborator·7 tasks done·

  ·

• feat: add multi-GPU backend abstraction layer (gpu_backend)

  #131·feiqiangs opened Mar 31, 2026Collaborator·

  ·

• feat: add MUSA support for FlexKV

  #126·superleo opened Mar 23, 2026·

  ·
  1

• add bypass cpu capability for local transfer

  #104·didoteebin opened Feb 9, 2026·

  ·
  1