80 Commits
Author SHA1 Message Date
MOVZX e0fe5df731 Merge branch 'main' into patches 2026-10-17 08:06:56 +00:00
Omar MuhammadNicholas SusantoDouglas Lehrmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
f019e3e105 [ROCm] attn res + sigmoid_mul + conv fusions (#52968)
Signed-off-by: Omar Muhammad <omar.muhammad@amd.com>
Signed-off-by: Nicholas Susanto <nicholas.susanto@amd.com>
Co-authored-by: Nicholas Susanto <nicholas.susanto@amd.com>
Co-authored-by: Douglas Lehr <91553416+dllehr-amd@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-10-17 00:54:54 +00:00
28e92eeff2 [ROCm][Perf][Kimi-K3] Use AITER Gluon chunk_kda for KDA prefill (#59552)
Signed-off-by: Omar Muhammad <omar.muhammad@amd.com>
Signed-off-by: Nicholas Susanto <nicholas.susanto@amd.com>
Co-authored-by: Nicholas Susanto <nicholas.susanto@amd.com>
Co-authored-by: Douglas Lehr <91553416+dllehr-amd@users.noreply.github.com>
Co-authored-by: Kevin H. Luu <khluu000@gmail.com>
2026-10-16 22:58:56 +00:00
Oxana KorzhandClaude Opus 5.5 5bf3d9f665 [ROCm][Kimi-K3] Support DSpark adaptive verification on MI355X (#59824)
Signed-off-by: Oxana Korzh <okorzh@amd.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
2026-10-16 13:53:49 -05:00
MOVZX be95edb7f6 Merge branch 'main' into patches
Signed-off-by: MOVZX <movzx@yahoo.com>
2026-10-16 17:35:42 +07:00
3f1e57225a [ROCm][Perf][Kimi-K3] Use AITER Gluon KDA kernels for decode and spec decode (#57392)
Signed-off-by: omuhamma <omar.muhammad@amd.com>
Signed-off-by: Nicholas Susanto <nicholas.susanto@amd.com>
Co-authored-by: Nicholas Susanto <nicholas.susanto@amd.com>
Co-authored-by: Simon Danielsson <70206058+simondanielsson@users.noreply.github.com>
2026-10-15 17:54:24 -05:00
MOVZX 76df77c04e Merge branch 'main' into patches
Signed-off-by: MOVZX <movzx@yahoo.com>
2026-10-15 21:16:39 +07:00
Rohan PotdarandClaude Opus 5.5 52a8cc5316 [ROCm][CI] Add dtype to the KDA metadata test's fake model config (#61650)
Signed-off-by: Rohan138 <rohanpotdar138@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-14 18:04:14 +00:00
Andreas KaratzasandClaude 67516b6c42 [Bugfix][ROCm][Kimi-K3] Stage fused KDA decode weights after loading (#61207)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
2026-10-13 14:55:57 -05:00
c5bda658d2 [Model][MoE] Load MegaMoE experts directly into the DeepGEMM layout (#60096)
Signed-off-by: aoshen02 <aoshen@inferact.ai>
Signed-off-by: Aaron Hao <hao-aaron@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Aaron Hao <hao-aaron@users.noreply.github.com>
2026-10-13 11:52:08 -07:00
dingdangmaoandCodex 124b89d5b7 [Bugfix][RecoverSSM] Correct aligned state indices at exact block boundaries (#59962)
Signed-off-by: apex-mochen <2756823972@qq.com>
Co-authored-by: Codex <noreply@openai.com>
2026-10-12 02:08:14 +00:00
Nick Hill c059bee812 [MRV2] Switch CI tests still running on MRV1 unnecessarily (#61014)
Signed-off-by: Nick Hill <nickhill123@gmail.com>
2026-10-11 09:57:07 -07:00
f8e6e365ae [Bugfix][Mooncake] Save the EAGLE attention proof with hybrid Mamba partial tails (#60695)
Signed-off-by: Dao Le <Dao007forever@gmail.com>
Signed-off-by: Dao007forever <daole@inferact.ai>
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: zjy0516 <riverclouds.zhu@qq.com>
2026-10-10 00:08:08 +00:00
MOVZX ac305c44cb Merge branch 'main' into patches
Signed-off-by: MOVZX <movzx@yahoo.com>
2026-10-09 17:41:13 +00:00
Rohan PotdarandClaude Opus 5.5 88a419649e [ROCm][Kimi-K3] Use vLLM's FP8 max in the fused AttnRes quantization (#60717)
Signed-off-by: Rohan138 <rohanpotdar138@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-09 04:45:13 +00:00
Wei ZhaoWei ZhaoCodexzjy0516Claude Opus 5.5mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
d8fa463e2b [Bugfix][Prefix caching] Fix inconsistent hybrid Mamba prefix cache boundaries (#60533)
Signed-off-by: Wei Zhao <weizha@aws-cmh-slurm-1-login-01.cm.cluster>
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: Wei Zhao <weizha@aws-cmh-slurm-1-login-01.cm.cluster>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-10-08 18:33:48 +00:00
Elvir CrnčevićandMisha Goin 8674a794c7 [Perf] Enable Kimi K3 shared expert sharding with DeepEP v2 (#56875)
Co-authored-by: Misha Goin <mgoin64@gmail.com>
2026-10-08 16:42:15 +00:00
9f3d70e081 [Quantization] Enable shared expert fusion compatibility with online shared_expert quantization (showcase: along Quark MXFP4 routed experts) (#55686)
Signed-off-by: Felix Marty <Felix.Marty@amd.com>
Signed-off-by: fxmarty-amd <felmarty@amd.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
Co-authored-by: Misha Goin <mgoin64@gmail.com>
2026-10-08 12:04:46 -04:00
kliuae d80ddaac73 [ROCm][Perf] Kimi-K3 Store only the current rank's shards in latent MoE up-proj (#59591)
Signed-off-by: kliuae <kuanfu.liu@embeddedllm.com>
2026-10-08 14:55:20 +00:00
458ba2edf8 [ROCm][Kimi-K3] Fuse AttnRes output with per-token FP8 input quantization (#59069)
Signed-off-by: whx-sjtu <xiaowang990929@gmail.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: Fangzhou Ai <31551580+Fangzhou-Ai@users.noreply.github.com>
2026-10-08 12:01:47 +00:00
b2863c0de9 [Kimi-K3] Use the fused K/V pack kernel on the decode-context-parallel prefill path (#56881)
Signed-off-by: Brandon Zhang <31413216+brandonfzhang@users.noreply.github.com>
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
Co-authored-by: Wei Zhao <51183510+wzhao18@users.noreply.github.com>
2026-10-06 00:13:55 +00:00
Vadim GimpelsonCursor Agentmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
0eac152707 [Bugfix] Route every speculation-capable row through the speculative path so recurrent state stays consistent (#56531)
Signed-off-by: Vadim Gimpelson <vadim.gimpelson@gmail.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-10-06 03:03:44 +04:00
MOVZX 85c31d47ad Merge remote-tracking branch 'upstream/main' into patches
Signed-off-by: MOVZX <movzx@yahoo.com>
2026-10-03 22:06:10 +07:00
MOVZX 2877121b2d Merge remote-tracking branch 'upstream/main' into patches
Signed-off-by: MOVZX <movzx@yahoo.com>
2026-10-03 21:59:03 +07:00
Markus Hartikainen b0e21b3083 [ROCm][Kimi-K3] Add opt-in gfx942 MXFP4-to-int4 conversion (#51274)
Signed-off-by: Markus Hartikainen <markus.hartikainen@amd.com>
Signed-off-by: Hartikainen <mahartik@amd.com>
2026-10-03 14:28:02 +08:00
Jiangyun ZhuandClaude Opus 5.5 d648847fe4 [CI][Kimi-K3] Test prefix cache reuse with KV offload, P/D and DCP (#59229)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-02 20:58:35 +08:00
kliuae 0154979656 [ROCm][Perf] Kimi-K3 enable prefill checkpoints on ROCm (#58344)
Signed-off-by: kliuae <kuanfu.liu@embeddedllm.com>
2026-10-02 06:58:04 +00:00
Duncan MossRoger Wangmergify[bot] <37929162+mergify[bot]@users.noreply.github.com>Thien Tran
766b4ae54c [Kimi-K3] Add FlashInfer speculative KDA backend (#54255)
Signed-off-by: Duncan Moss <dmoss@nvidia.com>
Co-authored-by: Roger Wang <hey@rogerw.io>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: Thien Tran <gau.nernst@yahoo.com.sg>
2026-10-01 02:28:50 +00:00
MOVZXandHinata ae0350c0de [Spec Decode][MRV2] RecoverSSM for Qwen GDN and PLE short conv (PR #58863)
Intake of open upstream PR vllm-project/vllm#58863. Active only with
VLLM_USE_V2_MODEL_RUNNER=1; inert otherwise.

Speculative decoding on Gated DeltaNet snapshotted the full recurrent
state after every draft position and reserved one Mamba block per draft
position. On a 24 GB card that is reserved VRAM we never use, plus
per-step copy traffic.

RecoverSSM, the protocol Kimi-K3 KDA already uses, replaces this:
verify runs from a read-only checkpoint and stores a small per-token
replay record [correction (V), normalized key (K), gate (1)] in fp32;
commit replays only the accepted tokens and writes the recurrent state
once, plus the block-boundary state in align mode. The PLE dilated
short conv compacts its accepted window with the same plan.

Conflict in vllm/models/qwen4_exp/nvidia/mtp.py: our get_top_tokens()
work put LocalArgmaxMixin on Qwen4ExpMTP, the PR adds SupportsReplaySSM
to the same base list. Kept both.

Co-Authored-By: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>
2026-09-29 13:39:09 +07:00
xaguilar-amdandmergify[bot] <37929162+mergify[bot]@users.noreply.github.com> a7ff44355c [ROCm][Perf] Kimi-K3 Enable sharded latent MoE up-projection under EP (#54956)
Signed-off-by: Xavier Aguilar <xavier.aguilarfruto@amd.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
2026-09-28 10:31:29 -04:00
dd3bc9c72f [Perf][MRV2] Reuse Mamba/GDN metadata across KV cache groups (#58762)
Signed-off-by: Luca Motz <luca.motz@icloud.com>
Co-authored-by: Xin Yang <xyangx@amazon.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Lucas Wilkinson <LucasWilkinson@users.noreply.github.com>
2026-09-28 04:36:58 +00:00
5567cc1b25 [Mamba] RecoverSSM align: keep a boundary-ending state in its block
A speculative window whose accepted tokens end exactly on a recurrent block
boundary selected the following block-table column for the committed state
(commit plan) and recorded it as the running column (postprocess). That
column can still be unallocated, so the next step reads an uninitialized
page. Select the block that holds the last committed token, matching the
(n - 1) // block convention used elsewhere in align mode. Applies to
Kimi-K3 KDA and to the Qwen GDN/PLE RecoverSSM paths that share the plan.

Fix and test from local-inference-lab/vllm a02f585d (Martin Vit, root cause
reported by astra); the GDN/PLE align tests now expect the same convention.

Signed-off-by: Jürgen Schmied <juergenschmied70@gmail.com>
Co-authored-by: Martin Vit <martin@voipmonitor.org>
Co-authored-by: Claude <noreply@anthropic.com>
2026-09-27 11:12:40 +02:00
Jürgen SchmiedandClaude a80e41d690 [Mamba] Share RecoverSSM commit kernels and helpers across models
RecoverSSM's commit-plan and conv-window compaction kernels are model independent.
Move them from models/kimi_k3 into model_executor/layers/mamba/ops/recoverssm.py
(Kimi-K3 keeps its private names as import aliases; kernel source unchanged), add
mamba/recoverssm_utils.py (activation predicate, argument validation) and the
generic spec-row/request mapping helpers to v1/attention/backends/recoverssm_metadata.py.

CacheConfig.use_kda_recoverssm becomes use_recoverssm: the switch now selects RecoverSSM
for any model whose recurrent layers implement it, not only KDA.

No functional change for Kimi-K3.

Signed-off-by: Jürgen Schmied <juergenschmied70@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>
2026-09-27 00:07:04 +02:00
1417022c3d [ROCm][Kimi-K3] Optimize low-concurrency speculative KDA (#58045)
Signed-off-by: jiacao-amd <jiahui.cao@amd.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>
2026-09-25 12:33:30 -05:00
Yongye ZhuandClaude Opus 5.5 378504a544 [MoE] Defer the TRTLLM-Gen top-k finalize on the modular path (#58635)
Signed-off-by: Yongye Zhu <zyy1102000@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-25 15:38:29 +08:00
c4cfd7007b [CI][ROCm] Add an MI355 Kimi-K3 unit test group (#58012)
Signed-off-by: Oxana Korzh <okorzh@amd.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>
2026-09-23 11:25:51 -05:00
382970ee6c [Kimi-K3][AMD] Return KDA and MLA projection outputs directly (#50592)
Signed-off-by: Liuyinfeng01 <yinfeliu@amd.com>
Co-authored-by: Liuyinfeng01 <199041580+LiuYinfeng01@users.noreply.github.com>
Co-authored-by: Andreas Karatzas <akaratza@amd.com>
2026-09-21 15:44:41 -05:00
Chaunceyandzjy0516 3df4ae153e [Core][KDA] Generalize Mamba prefill checkpoint builder and exporter (#57783)
Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: zjy0516 <riverclouds.zhu@qq.com>
2026-09-21 04:46:11 +00:00
Harry MellorandClaude Opus 5 651a88c09a [CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-16 10:57:24 -07:00
Harry MellorandClaude Opus 5 8c1557a79c Add pydocstyle to the ruff rules (#52136)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Signed-off-by: Harry Mellor <hej.mellor@gmail.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-16 16:44:29 +00:00
sashko-zakharchukandzjy0516 f8b5c11468 [Bugfix][Kimi-K3] Do not classify a stateless first chunk as a decode (#51483)
Signed-off-by: Oleksandr Zakharchuk <oleksandr.zakharchuk@gmail.com>
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: zjy0516 <riverclouds.zhu@qq.com>
2026-09-16 21:24:48 +08:00
Wei Zhao 9d88ceb026 [KDA] Update flashKDA to support bf16 checkpoint state (#56485)
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
2026-09-11 23:53:24 +00:00
siyuandIsotr0py 73fb19151f [Fast Start] Support fp4 (#55465)
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
2026-09-10 13:22:17 +00:00
Wentao Ye 86aca66191 [Kimi K3 Perf] Avoid KDA mixed-batch gather/scatter, 5.2%~7.7% E2E Throughput Improvement (#56159)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
2026-09-10 17:57:01 +08:00
9521c60bdc [ROCm][Perf] Kimi-K3 Fused kernels for KDA prefill reland (#54038)
Signed-off-by: kliuae <kuanfu.liu@embeddedllm.com>
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Signed-off-by: kliuae <17350011+kliuae@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
2026-09-10 15:37:54 +08:00
Sherif WalyandNicolò Lucchesi f998862d46 [Bugfix] Fix Kimi K3 loading with interleaved weight streams (#53379)
Signed-off-by: Sherif Waly <sherif.waly@mistral.ai>
Co-authored-by: Nicolò Lucchesi <nicolo.lucchesi@mistral.ai>
2026-09-08 16:08:51 +02:00
Wei Zhao a69402aaa8 [Perf] Integrate FlashInfer KDA kernels (#55364)
Signed-off-by: wzhao18 <wzhao18.sz@gmail.com>
2026-09-07 20:29:30 -07:00
Ziming HuangandOpenAI Codex 144e79c810 [Kimi K3] Support internal prefix checkpoints with partial prefix caching and spec-decoding (#53614)
Signed-off-by: Ziming Huang <zelda.huanghuang@gmail.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
2026-09-06 13:18:15 +08:00
d87a440f88 [Core][MRV2] Support eagle3 spec decode with pipeline parallel (#50514)
Signed-off-by: Yongqin Wang <yongqinwang@roblox.com>
Co-authored-by: Cursor <noreply@cursor.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
2026-09-05 11:45:34 +08:00
6cbb3c154e [Perf][GDN] Build cudagraph-capture metadata without a device sync (#55404)
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Co-authored-by: Roger Wang <rogerw@inferact.ai>
Co-authored-by: Kimi Code <noreply@moonshot.cn>
2026-09-04 17:40:17 -07:00