All pull requests

Pull requests

• [Test] Cover cross-layer KV binding through patched reshape

  #508·vmanvs opened Sep 22, 2026·

  ·

• fix(sglang): evict prefix-cache blocks by page

  #507·jeff3071 opened Sep 22, 2026·

  ·

• fix: reject ambiguous worker device indices

  #503·shipiyouniao opened Sep 18, 2026·

  ·
  1

• feat: Intel XPU (Level Zero) backend for kvcached VMM

  #501·kpjeeja opened Sep 17, 2026·

  ·
  29

• fix(vllm): bind V1 hybrid Mamba pages on 0.28

  #500·shipiyouniao opened Sep 17, 2026·

  ·
  1

• ci: automate bounded vLLM release compatibility checks

  #496·shipiyouniao opened Sep 15, 2026·

  ·

• tools: add bounded Codex engine compatibility repair

  #495·shipiyouniao opened Sep 15, 2026·

  ·
  1

• feat(vllm): integrate attention KV storage with Model Runner V2

  #494·shipiyouniao opened Sep 15, 2026·

  ·
  1

• feat(vllm): support 0.28 packed attention KV storage

  #493·shipiyouniao opened Sep 15, 2026·

  ·
  1

• fix(vllm): preserve profiling state for 0.28 warmup

  #492·shipiyouniao opened Sep 15, 2026·

  ·
  2

• feat: allocate physical KV pages in request-sized batches

  #489·shipiyouniao opened Sep 14, 2026·

  ·

• perf: coalesce physical growth retries and signal released capacity

  #488·shipiyouniao opened Sep 14, 2026·

  ·

• feat: add native physical KV reservation APIs

  #487·shipiyouniao opened Sep 14, 2026·

  ·

• fix(vllm): unlink the /dev/shm segment when EngineCore shuts down

  #482·rishabhsinha17 opened Sep 2, 2026·

  ·
  1

  15

• fix: ship kvcached_autopatch.pth in editable installs and warn when it is missing

  #481·rishabhsinha17 opened Sep 2, 2026·

  ·
  1

  4

• feat: lifecycle readiness and failure propagation for KVCacheManager (#375 item 5, phase 1)

  #478·rishabhsinha17 opened Sep 2, 2026·

  ·
  20

• fix: split page create from map to avoid a remap race on Volta

  #465·andrewleech opened Aug 26, 2026·

  ·
  3

• fix(vllm): defer physical unmap until async batches complete

  #464·shipiyouniao opened Aug 23, 2026·

  ·
  1

  3

• docs: explain revisioned memory limit integration

  #458·shipiyouniao opened Aug 21, 2026·

  ·
  7

• fix: coordinate distributed physical KV mapping outcomes

  #455·shipiyouniao opened Aug 19, 2026·

  ·
  9

  

• Add page-granular CPU offloading foundation

  #450·Lanoxia opened Aug 17, 2026·

  ·

• Add manually triggered model compatibility matrix

  #440·Lanoxia opened Aug 10, 2026·

  ·
  3

• Fix vLLM padded KV page allocation and strides

  #426·shipiyouniao opened Aug 5, 2026·

  ·
  1

  3

• Add automated vLLM and SGLang upstream synchronization

  #423·Lanoxia opened Aug 4, 2026·

  ·
  8

  

• Export kvcached metrics through SGLang's native endpoint

  #421·shipiyouniao opened Aug 3, 2026·

  ·
  8