v0.6.0

What's Changed

• [feat] Add H2D and D2H Path Based on GPUDirect RDMA by @relat-ivity in #958
• [feat] Support UCM hybrid KV cache connector for mixed Mamba/Attention block layouts by @wangwenxin0312 in #957
• [bugfix] buffer size & request_finished_all_groups fix by @wangwenxin0312 in #963
• [fix] Update patch to track mindie changes by @yumingyue624 in #909
• [CI] Add vLLM Dockerfiles to support deepseek v4 by @dante159753 in #959
• [opt]opt hma by @wuhuxiao in #968
• [feat] Adapt UCM connector for latest vLLM-Ascend updates by @wangwenxin0312 in #964
• [feat]Add Cache|Fake in pipeline by @qyh111 in #961
• [bugfix] Split SFA patch by vllm-ascend version and fix patch runtime issues by @qyh111 in #975
• [opt] Fix FAWA HMA connector dump handling and optimize MLA TP dump balance. by @wuhuxiao in #976
• [bugfix] Fix FAWA async dump task block release by @wuhuxiao in #986
• [bugfix]Fix update_graph_params by @qyh111 in #983
• Add layerwise & pipeline store metrics to monitor performance by @dante159753 in #960
• [Feat] Restore Ascend PC e2e gate by @dante159753 in #994
• clean: fix clean code issues in compress_lib and related files by @Fencyee in #1004
• [Feat]Wa dump blockwise & support for vllm-ascend:0.20.2rc1 by @wuhuxiao in #992
• [Docs] Update PipelineStore parameters and quickstart patch docs, add… by @sumingZero in #1005
• [Feat] Logger redesign by @dante159753 in #1015
• [Bugfix] fix dump error when load failure by @flesher0813 in #1007
• [feat] Add patch support and required connector interfaces by @wangwenxin0312 in #1016
• [opt]Defer UCM KV Dump Waiting Until Request Completion by @qyh111 in #989
• [OPT] Support vLLM 0.20.2 runtime patching for UCM/HMA KV load-failure recovery. by @wuhuxiao in #1026
• [fix]clean code fix by @qyh111 in #1020
• add mindie compile files by @snoopyri in #1029
• [Bugfix] Enforce strict timeout for Posix AIO by @dante159753 in #1022
• [Opt] check / load task error handle by @wuhuxiao in #1035
• Modify idea.png by @qyh111 in #1036
• fix(compress): resolve clean code warnings and fix memory pool bugs by @Fencyee in #1019
• [BugFix] Dump last mtp decode result by @flesher0813 in #1037
• [Fix] add patch fixing SWA block double free && Add fawa metric by @wuhuxiao in #1043
• [Feat] Integrate ucm metrics into vLLM connector metrics by @dante159753 in #995
• feat (calculator): Added a hybrid model tab and support V4 by @Potterluo in #996
• fix(compress): remove securec library to resolve open-source complian… by @Fencyee in #1046
• [Opt] change load_blocks_threshold to load_tokens_threshold and fix duplicated block free by @wuhuxiao in #1045
• Add Ascend A3 Cache SDMA Direct copy path by @NaganooMei in #1039
• [Feat] Mooncake|Posix by @UESTC-AHao in #972
• [bugfix]bugfix for duplicate free by @qyh111 in #1053
• [Bugfix] Select SDMA direct launch granularity by connector mode by @NaganooMei in #1057
• [Bugfix] Remove embedded loader path settings by @dante159753 in #1058
• Fix Ascend hybrid prefix cache truncation for compressed attention groups by @qyh111 in #1059
• [Bugfix] Environment pre-check code adaptation by @Menglths in #1006
• [Feat] Add install.sh to support custom post install actions by @dante159753 in #1044
• disable migratepage by @snoopyri in #1065
• [Fix] Remove Register_Memory in StoreV1 by @UESTC-AHao in #1070
• [fix]Add UCMConnectorV1 in factory by @qyh111 in #1067
• fix: recompute enough tokens on full UCM cache hit by @NaganooMei in #1068
• [Feat] Add UCM connector and Mooncake metrics dashboards by @dante159753 in #1066
• [Doc] Add Mooncake Doc by @UESTC-AHao in #1064
• [Feat]Add ucm-toolkit: unified CLI for dev-sandbox, posix-aio, nic-monitor, and metrics-view by @harrisonyhq in #1012
• [BugFix] Update the hotness of all ranks by @flesher0813 in #1071
• Codex/add sandbox copy cases by @NaganooMei in #1077
• [docs] add supported models by @Lijiachen1018 in #1072
• [fix] fix llmperf by @Lijiachen1018 in #1080
• fix: align UCM and vllm-ascend CPU affinity by @snoopyri in #1074
• switch version by @qyh111 in #1085

New Contributors

• @relat-ivity made their first contribution in #958
• @snoopyri made their first contribution in #1029

Full Changelog: v0.5.0...v0.6.0