Commit Graph
479 Commits
Author SHA1 Message Date
Sihan ChenandLi, Jiang <jiang1.li@intel.com> 4938d44a3b [CPU] fixes heterogeneous NIXL KV transfer into CPU_ATTN decode workers (#47871)
Signed-off-by: Spycsh <sihan.chen@intel.com>
Co-authored-by: Li, Jiang <jiang1.li@intel.com>
2026-07-20 07:33:13 +00:00
b6ff8a2f50 [Core] Add MRV2 virtual-batch PCP for MLA (#46570)
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
2026-07-19 02:53:15 +00:00
AlexHuangandOr Ozeri f38f3d11fb [Bugfix][KV Offloading] Offload last block at request finish and prevent reuse race (#48596)
Signed-off-by: Alex <alex.tech.lab@outlook.com>
Co-authored-by: Or Ozeri <oro@il.ibm.com>
2026-07-17 16:50:49 +03:00
426d48bfa1 [KV Offload] Add optional tier locality to FS/OBJ KV events (#48281)
Signed-off-by: Change72 <changg@nvidia.com>
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-17 10:31:52 +03:00
Debasish MohantyandOr Ozeri 472d330c21 Add blocks_per_chunk configuration for KV offloading to support heterogeneous KV cache groups (#48878)
Signed-off-by: Debasish-87 <22btics06@suiit.ac.in>
Co-authored-by: Or Ozeri <oro@il.ibm.com>
2026-07-17 09:00:12 +03:00
971dac2caa [Bugfix][KV-transfer] MoRIIO: retry RDMA send-queue-full backpressure instead of failing the read (#47495)
Signed-off-by: Edwin Lim <edwin.lim@mangoboost.io>
Signed-off-by: Edwin Lim <edwinlim0919@gmail.com>
Signed-off-by: harishk-mangoboost <harish.kambhampaty@mangoboost.io>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: harishk-mangoboost <harish.kambhampaty@mangoboost.io>
2026-07-16 20:02:27 +00:00
Nicolò Lucchesi a317bc5739 [Misc][Nixl] Unify _logical_to_remote_kernel_block_ids (#48717)
Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai>
2026-07-16 18:36:25 +08:00
a9531edfa6 [KV Offload] Define clean backend configuration boundary (#48150)
Signed-off-by: Change72 <changg@nvidia.com>
Signed-off-by: Chang Guo <cguo51@asu.edu>
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Or Ozeri <or@ozery.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
2026-07-16 13:27:05 +03:00
Ilia Yastrebov 9f8cbfd8eb Vectorize prep xfer list creation (#48209)
Signed-off-by: Ilia Yastrebov <iyastrebov@nvidia.com>
2026-07-16 11:39:05 +02:00
Elvir Crnčević 12f2c515a7 [Bugfix] Fix offloading set_ overflow for packed non-uniform KV caches (#48530)
Signed-off-by: Elvir Crncevic <elvircrn@gmail.com>
2026-07-16 10:04:39 +03:00
Nicolò Lucchesi 66b6c684ab [PD][Bugfix] Fix validation of cache shape for attn backends enforcing different kernel_block_size (#48125)
Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai>
2026-07-15 18:26:02 +08:00
32aef44388 [Bugfix] Include inline per-token-head scales in offloaded page transfer width (#48411)
Signed-off-by: Itay Etelis <itay.etelis@ibm.com>
Signed-off-by: Itay Etelis <Itay.etelis@gmail.com>
Signed-off-by: Itay Etelis <92247226+Etelis@users.noreply.github.com>
Co-authored-by: Itay Etelis <itay.etelis@ibm.com>
Co-authored-by: Itay Etelis <Itay.etelis@gmail.com>
2026-07-14 16:07:26 +03:00
tomerg-nvidiaandNickLucche 7a74a9662b [NIXL] Avoid reading expired blocks in bidirectional turn-2 read (#47021)
Signed-off-by: Tomer Gilad <tgilad@nvidia.com>
Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai>
Co-authored-by: NickLucche <nicolo.lucchesi@mistral.ai>
2026-07-14 13:03:41 +00:00
Nick Hill 8ac8375270 [Core] Preserve Marconi caching with selective hybrid cache retention (#47782)
Signed-off-by: Nick Hill <nickhill123@gmail.com>
2026-07-13 21:24:20 +01:00
AlexHuang 4c81772e8b [Bugfix][KV Offloading] Fix stale transfer_jobs after reset_cache + harden job completion (#48102)
Signed-off-by: Alex <alex.tech.lab@outlook.com>
2026-07-12 20:00:04 +03:00
Jiangyun ZhuandYifan Qiao 481e481be7 [2/N][Core] support partial prefix cache hit for hybrid model (#46384)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai>
Co-authored-by: Yifan Qiao <yifanqiao@inferact.ai>
2026-07-12 05:37:51 +00:00
51878e5b6e [2/N][KV-Cache Layout Refactor] Pack K/V into the content dim across attention backends (#44455)
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Signed-off-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
Co-authored-by: Matthew Bonanni <mbonanni@redhat.com>
Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com>
2026-07-11 11:11:16 -04:00
2ded1b24e7 [KV Connector][Mooncake] Apply SWA lookup mask before hashing/key build (#47317)
Signed-off-by: Zhewen Li <zhewenli@inferact.ai>
Co-authored-by: Zhewen Li <zhewenli@inferact.ai>
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: Claude <noreply@anthropic.com>
2026-07-09 19:51:23 +00:00
weishu 2285cfca46 [KVConnector] MultiConnector: give every sub-connector the request's real blocks in update_state_after_alloc (#46865)
Signed-off-by: deng451e <838677410@qq.com>
2026-07-09 11:10:39 -07:00
qizixiandNick Hill 7802c20c4e [KVConnector][NIXL] Support pipeline-parallel prefill in push mode (#45880)
Signed-off-by: zixi-qi <zixi@inferact.ai>
Co-authored-by: Nick Hill <nickhill123@gmail.com>
2026-07-08 16:49:23 -07:00
3354dba381 [Bugfix][KV offload] Store interior chunk-boundary blocks under MTP/Eagle (#46972)
Signed-off-by: Mikhail Kostryukov <mike@triptrack.net>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Or Ozeri <oro@il.ibm.com>
2026-07-07 13:16:52 +03:00
e040899a00 [KV Offloading] Add basic offloading metrics (#45958)
Signed-off-by: srinivas_oo7 <sklinkedin0120@gmail.com>
Signed-off-by: Srinivas Krovvidi <194645829+Srinivasoo7@users.noreply.github.com>
Co-authored-by: srinivas_oo7 <sklinkedin0120@gmail.com>
Co-authored-by: Or Ozeri <or@ozery.com>
Co-authored-by: Or Ozeri <oro@il.ibm.com>
2026-07-07 09:26:28 +03:00
Kunshang JiandHarry Mellor 07f9baf756 Revert "[Platform] Replace torch.cuda.Event with torch.Event (#47140)" (#47668)
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
2026-07-06 14:18:33 +01:00
Kunshang JiandHarry Mellor e840f0d3f5 [Platform] Replace torch.cuda.Event with torch.Event (#47140)
Signed-off-by: Kunshang Ji <kunshang.ji@intel.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
2026-06-30 08:39:59 -07:00
d8f483dc30 [Spec Decode] Fix hidden-state extraction block size for hybrid verifiers (#46301)
Signed-off-by: Igor Margulis <igor.margulis@intel.com>
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: mgoin <mgoin64@gmail.com>
2026-06-30 08:19:51 -07:00
Ronen SchafferandOr Ozeri 0fc2512094 [KV Offload] Pass ScheduleEndContext to on_schedule_end hook (#46450)
Signed-off-by: Ronen Schaffer <ronen.schaffer@ibm.com>
Co-authored-by: Or Ozeri <oro@il.ibm.com>
2026-06-30 17:07:12 +03:00
Dakai An 0feca7ffa8 PD disagg with Mooncake Connector: GDN support (Qwen3.5) and MLA support (Deepseek-V4-Flash) (#46807) 2026-06-29 23:29:04 -07:00
weishu 77654d080c [KVTransfer] MultiConnector: merge kv_transfer_params dicts across connectors (#46777)
Signed-off-by: deng451e <838677410@qq.com>
2026-06-30 00:25:05 +00:00
Wei Zhao 95528527ea [Bugfix][Mooncake] Fix Mooncake lookup prefixes with DCP > 1 (#46855)
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
2026-06-28 14:36:23 -07:00
Varun Sundar RabindranathandVarun Sundar Rabindranath 798185d438 [KV-Offloading] Fix tensors_per_block stride (#46888)
Signed-off-by: <>
Co-authored-by: Varun Sundar Rabindranath <varun-sundar-rabindranath@h100-01.nemg-001.lab.rdu2.dc.redhat.com>
2026-06-27 21:01:45 -04:00
Asaf Gardinandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> 15be78732b [NIXL][Mamba] Add Mamba1 support to NIXL P/D disaggregation (#45019)
Signed-off-by: Josephasafg <ajgard7@gmail.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-06-25 05:50:41 -07:00
Yifan QiaoClaudemergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
9e88e969c0 [Perf][KVConnector][Mooncake] Parallelize KV load with a receive-thread pool (#45971)
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-06-24 18:25:12 -07:00
Yifan Qiao b69816043a [Bugfix][MooncakeStore] track resumed requests via scheduler's resumed_req_ids (#46595)
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai>
2026-06-24 23:50:56 +00:00
Lucas Wilkinson e7df232288 [KV Offload] Gate packed HMA KV cache on cross-layer config (#46252)
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
2026-06-24 11:55:30 -04:00
Ronen SchafferandOr Ozeri bb61177e49 [KV Offloading] Replace bool|None lookup return with LookupResult enum (#46363)
Signed-off-by: Ronen Schaffer <ronen.schaffer@ibm.com>
Co-authored-by: Or Ozeri <oro@il.ibm.com>
2026-06-24 18:06:08 +03:00
Nicolò Lucchesi a2cb08b3d5 [Misc][PD] Disable bidirectional xfer mode for NixlPushConnector (#46473)
Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai>
2026-06-24 21:14:05 +08:00
Rui YinandOr Ozeri cf9fd6457e Fix KV offload request-finished lifecycle contract (#46284)
Signed-off-by: test test <2260891073@qq.com>
Signed-off-by: Rui Yin <2260891073@qq.com>
Co-authored-by: Or Ozeri <oro@il.ibm.com>
2026-06-24 15:42:40 +03:00
Martin Hickey f237e16b41 [KV Offload] Replace OffloadingHandler with OffloadingWorker (#45053)
Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
2026-06-24 14:44:24 +03:00
Wei ZhaoandYifan Qiao d20dbf921b [Mooncake] Only check and store new KV cache range (#46412)
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai>
Co-authored-by: Yifan Qiao <yifanqiao@inferact.ai>
2026-06-24 03:10:50 -07:00
Wei Zhao 05a0caba91 [Mooncake] Optimize lookup pool key string construction (#46188)
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
2026-06-24 11:51:53 +08:00
Andreas Karatzas ceae5bcbda [ROCm][CI] Fix nixl tests (#45219)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
2026-06-23 13:11:40 -05:00
d32575a2d2 [ROCm][P/D] Support MoRIIO heterogeneous TP fan-in (#46332)
Signed-off-by: Tan Pin Siang <tanpinsiang@gmail.com>
Co-authored-by: vllmellm <vllm.ellm@embeddedllm.com>
Co-authored-by: Hongxia Yang <hongxia.yang@amd.com>
Co-authored-by: Jun Kang Chow <junkangchow@gmail.com>
Co-authored-by: Chun Fang <chun.fang@amd.com>
Co-authored-by: TianDi101 <ditian12@amd.com>
Co-authored-by: functionstackx <47992694+functionstackx@users.noreply.github.com>
Co-authored-by: tjtanaa <tunjian.tan@embeddedllm.com>
2026-06-23 10:33:23 +00:00
7e47fb72b5 [ROCm][P/D] Fix MoRIIO WRITE mode for mixed KV layouts (#46290)
Signed-off-by: Tan Pin Siang <tanpinsiang@gmail.com>
Co-authored-by: vllmellm <vllm.ellm@embeddedllm.com>
Co-authored-by: Hongxia Yang <hongxia.yang@amd.com>
Co-authored-by: Jun Kang Chow <junkangchow@gmail.com>
Co-authored-by: Chun Fang <chun.fang@amd.com>
Co-authored-by: TianDi101 <ditian12@amd.com>
Co-authored-by: functionstackx <47992694+functionstackx@users.noreply.github.com>
2026-06-23 12:12:51 +08:00
Dao007foreverandClaude Opus 4.8 3ce15fd574 [v1][kvconnector] DecodeBenchConnector: fill list/tuple (Mamba/KDA) KV caches (#45080)
Signed-off-by: Dao Le <Dao007forever@gmail.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-22 11:54:00 -07:00
Change72andClaude a9f7b2d41c [feature][kv_offload] Self-describing KV events for OffloadingConnector (#43468)
Signed-off-by: Change72 <changg@nvidia.com>
Co-authored-by: Claude <noreply@anthropic.com>
2026-06-22 07:27:46 +00:00
Srinivas Krovvidiandsrinivas_oo7 c441ad1c07 [KV Offloading] Add labeled metrics support (#45957)
Signed-off-by: srinivas_oo7 <sklinkedin0120@gmail.com>
Co-authored-by: srinivas_oo7 <sklinkedin0120@gmail.com>
2026-06-21 18:04:01 +00:00
junkang1991Hongxia YangTan Pin SiangvllmellmChun FangTianDi101functionstackxtjtanaamergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
b91b7726e0 [ROCm][P/D] Support MiniMax-M3 mixed KV layouts in MoRIIO READ mode (#46039)
Signed-off-by: Jun Kang Chow <junkangchow@gmail.com>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: Hongxia Yang <hongxia.yang@amd.com>
Co-authored-by: Tan Pin Siang <tanpinsiang@gmail.com>
Co-authored-by: vllmellm <vllm.ellm@embeddedllm.com>
Co-authored-by: Chun Fang <chun.fang@amd.com>
Co-authored-by: TianDi101 <ditian12@amd.com>
Co-authored-by: functionstackx <47992694+functionstackx@users.noreply.github.com>
Co-authored-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-06-21 12:55:19 +00:00
Palaiologos1453 d3ad8e8bcd [Bugfix] Defer offload reads while transfers are pending (#46231)
Signed-off-by: test test <2260891073@qq.com>
2026-06-21 14:30:13 +03:00
Yifan Qiao ab7fcbdd5d [Perf][KVConnector][Mooncake] Compact chunk-hash keys and zero-copy lookup wire format (#45969) 2026-06-20 15:00:11 -07:00
cc22621b51 [KV Offload] Support packed HMA KV cache layout (#46205)
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
Co-authored-by: Tyler Michael Smith <tlrmchlsmth@gmail.com>
2026-06-20 21:19:40 +00:00