All pull requests

Pull requests

• docs: move the SGLang support matrix to the v0.5.20 validation state

  #526·rishabhsinha17 opened Oct 1, 2026Collaborator·

  ·

• fix(sglang): read the resolved mem_fraction_static on 0.5.17+

  #525·rishabhsinha17 opened Oct 1, 2026Collaborator·

  ·

• fix(sglang): serve zero-attention-layer pools natively

  #524·rishabhsinha17 opened Oct 1, 2026Collaborator·

  ·

• fix(vllm): align hybrid KV blocks to the kvcached page

  #522·RixinLiu opened Sep 30, 2026Collaborator·

  ·
  1

• fix(vllm): fail early on MRV2 kernel-block splits in contiguous layout

  #521·RixinLiu opened Sep 30, 2026Collaborator·

  ·
  3

• fix(sglang): reject incompatible pool modes before configuration

  #520·shipiyouniao opened Sep 30, 2026Collaborator·

  ·
  2

• fix: keep a same-name replacement's socket when an old listener stops 🤖🤖🤖

  #519·Momoyeyu opened Sep 27, 2026·

  ·
  8

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

  #508·vmanvs opened Sep 22, 2026·

  ·
  3

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

  #507·jeff3071 opened Sep 22, 2026Contributor·

  ·
  10

• fix: reject ambiguous worker device indices

  #503·shipiyouniao opened Sep 18, 2026Collaborator·

  ·
  1

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

  #501·kpjeeja opened Sep 17, 2026·

  ·
  30

• ci: automate engine release analysis and bounded compatibility repair

  #496·shipiyouniao opened Sep 15, 2026Collaborator·

  ·
  2

• tools: add bounded Codex engine compatibility repair

  #495·shipiyouniao opened Sep 15, 2026Collaborator·

  ·
  2

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

  #489·shipiyouniao opened Sep 14, 2026Collaborator·

  ·

• perf: coalesce physical growth retries and signal released capacity

  #488·shipiyouniao opened Sep 14, 2026Collaborator·

  ·

• feat: add native physical KV reservation APIs

  #487·shipiyouniao opened Sep 14, 2026Collaborator·

  ·

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

  #478·rishabhsinha17 opened Sep 2, 2026Collaborator·

  ·
  22

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

  #465·andrewleech opened Aug 26, 2026·

  ·
  3

• docs: explain revisioned memory limit integration

  #458·shipiyouniao opened Aug 21, 2026Collaborator·

  ·
  7

• fix: coordinate distributed physical KV mapping outcomes

  #455·shipiyouniao opened Aug 19, 2026Collaborator·

  ·
  9

  

• Add page-granular CPU offloading foundation

  #450·Lanoxia opened Aug 17, 2026Contributor·

  ·

• Add manually triggered model compatibility matrix

  #440·Lanoxia opened Aug 10, 2026Contributor·

  ·
  3

• Fix vLLM padded KV page allocation and strides

  #426·shipiyouniao opened Aug 5, 2026Collaborator·

  ·
  1

  3

• Add automated vLLM and SGLang upstream synchronization

  #423·Lanoxia opened Aug 4, 2026Contributor·

  ·
  8

  

• Export kvcached metrics through SGLang's native endpoint

  #421·shipiyouniao opened Aug 3, 2026Collaborator·

  ·
  8