All pull requests

Pull requests

• fix(toolkit): make precheck version detection more portable

  #1556·student-jhz opened Oct 24, 2026Contributor·

  ·
  1

• [Feat][Cache] Unify Ascend and CUDA on partitioned memfd cache

  #1554·NaganooMei opened Oct 23, 2026Contributor·

  ·

• [Refactor] Replace mamba copy-order patch with vllm-ascend 0.30 prepare_mamba_state_copy protocol

  #1553·sumingZero opened Oct 23, 2026Contributor·

  ·

• [Feat] Integrate ASU KVClient metrics with UCM

  #1546·Crz-Willis opened Oct 22, 2026Contributor·6 of 7 tasks·

  ·
  17

• [Feat][Test][Doc] Add Decode KV writeback for multi-turn Prefill reuse

  #1542·lhecust721 opened Oct 20, 2026·

  ·
  1

• [Bugfix] Fix GLM-5.3-Flash video placeholder timestamps for vLLM 0.30.0

  #1524·harrisonyhq opened Oct 19, 2026Contributor·

  ·

• [Bugfix] Adapt HLA Connector to new vLLM KV cache layouts for Qwen models

  #1522·wz-cloud opened Oct 18, 2026Contributor·

  ·

• Update supported VLLM ascend versions

  #1519·Potterluo opened Oct 15, 2026Contributor·

  ·

• [CI/CD] optimise pipelines

  #1518·Lijiachen1018 opened Oct 15, 2026Contributor·

  ·
  13

• [Tool] Extend cache performance benchmarks and metrics reports

  #1495·NaganooMei opened Oct 13, 2026Contributor·

  ·

• [Opt] Fix misleading FAWA buffer capacity error and add early resourc…

  #1485·student-jhz opened Oct 10, 2026Contributor·

  ·

• [CI] Extend release image disk cleanup timeout to 30 minutes

  #1476·Potterluo opened Oct 8, 2026Contributor·

  ·

• [feat] 支持在线切换UCM模式，关闭、lite模式、启动

  #1473·student-jhz opened Sep 30, 2026Contributor·

  ·

• [Opt] Refine KV client observability metrics and Grafana dashboard

  #1449·yuanzhg078 opened Sep 24, 2026Contributor·

  ·

• [Feat] Step-wise asynchronous KV cache loading

  #1442·CoooolHai opened Sep 22, 2026Contributor·

  ·
  3

• [feat] distribute MLA dumps across TP ranks

  #1434·sunyuecheng01 opened Sep 21, 2026Contributor·

  ·
  2

• fix(test): merge duplicate pytest_sessionfinish hook in conftest

  #1368·Anai-Guo opened Sep 13, 2026·

  ·

• [bugfix] Import dataclasses in test_uc_performance.py

  #1365·Anai-Guo opened Sep 13, 2026·

  ·

• [Toolkit][SGLang] Add offline model compatibility checker for UCM integration

  #1362·lc1314555 opened Sep 11, 2026Contributor·

  ·

• [Feat] Refresh UCM connector metrics while vLLM engines are idle

  #1353·dante159753 opened Sep 10, 2026Contributor·

  ·

• [docs] quickstart: remove Manual Git Patch option and move runtime en…

  #1349·student-jhz opened Sep 9, 2026Contributor·

  ·

• feat: reject kv_load_failure_policy=recompute for hybrid UCM connectors

  #1343·ljccsu opened Sep 8, 2026Contributor·

  ·

• docs: add UCM configuration parameters reference

  #1330·ljccsu opened Sep 4, 2026Contributor·

  ·
  1

• [Feat] Support Delegator DramStore pipeline

  #1319·pyxyzc opened Sep 1, 2026Contributor·

  ·

• [Bugfix] Reinitialize logger backends after fork

  #1313·dante159753 opened Aug 31, 2026Contributor·

  ·