Commit graph

  • f077d5a5f2 [build]: sync sglang submodule to 04653fa88f5ce4ea83632c4ad436343db6bc8324 auto/sync-sglang ovowei 2026-07-24 10:03:17 +00:00
  • a8062bfa7e
    release: bump version to 0.6.4 (#2105) main yyj 2026-07-23 22:42:28 +08:00
  • 924754a00b
    [feat](kt-kernel): end-to-end full-parameter and LoRA SFT(#2094) v0.6.4 yyj 2026-07-23 21:23:34 +08:00
  • d1a3ed8a30
    [perf]: add K2 RAWINT4 prefill mat-mat dispatch (#2080) yyj 2026-07-19 21:43:01 +08:00
  • 937f61c988
    [fix](kt-kernel): AVX-VNNI-256 RAWINT4 per-expert weight loading (#2092) Aaryan Mehta 2026-07-19 05:24:52 -04:00
  • def0f9313d
    fix(security): bind balance_serve scheduler ZMQ socket to loopback (#2087) (#2091) Tai An 2026-07-18 20:38:22 -07:00
  • 821f3b6198 deploy: 0c2912a5e2 gh-pages yyj6666667 2026-07-17 08:24:56 +00:00
  • 0c2912a5e2
    [feat] Add SYCL backend for intel iGPU (#2089) callmegaga 2026-07-17 16:24:04 +08:00
  • 01fdfa609e
    docs(DSV4): add supplementary deployment and tuning notes Graham 2026-07-15 00:06:58 -05:00
  • ef590428a1 deploy: 7c021b430c yyj6666667 2026-07-10 03:21:48 +00:00
  • 7c021b430c
    [chore](sglang): bump sglang submodule for FP16 FE8M0 Marlin fix (#2082) callmegaga 2026-07-10 11:18:45 +08:00
  • cb9f47d142
    [fix](cli): detect bound ports before launch (#2071) VectorPeak 2026-07-06 18:25:31 +08:00
  • 79b265b2f6
    fix: normalize compressed RAWINT4 weights (#2075) lutianshu824 2026-07-06 18:06:52 +08:00
  • d076e63b25
    Merge b1b2174a87 into 8e46e5896c Shaurya Omar 2026-06-25 20:38:30 +00:00
  • b1b2174a87 feat(kt-kernel): add score-aware expert placement Shaurya Omar 2026-06-26 01:39:42 +05:30
  • 8e46e5896c
    release: bump version to 0.6.3.post1 (#2063) v0.6.3.post1 Benjamin 2026-06-25 22:42:55 +08:00
  • 99cf022b0a release: bump version to 0.6.3.post1 yyj 2026-06-25 22:38:00 +08:00
  • 93078f8231 ci(release-pypi): set up Python in publish-pypi job (fixes PEP 668) yyj 2026-06-25 22:18:05 +08:00
  • 0e73671b18
    Merge 97c033d817 into 459df61262 devangpratap 2026-06-25 12:46:29 +00:00
  • 531cbcbf39
    Merge 48b628c358 into 459df61262 sysy 2026-06-25 12:46:28 +00:00
  • 459df61262
    [build] Sync sglang-kt submodule (v0.6.3) (#2055) github-actions[bot] 2026-06-25 20:41:04 +08:00
  • 4317bff7f8 ci(release-pypi): target manylinux_2_35 (build host glibc), drop dishonest fallback yyj 2026-06-25 20:31:53 +08:00
  • f8fd344610 ci(release-pypi): verify wheel inside $RUNNER_TEMP, not /tmp yyj 2026-06-25 16:59:32 +08:00
  • c02eac80a9 ci(release-pypi): skip apt-get on self-hosted runners; split pip indexes yyj 2026-06-25 15:59:15 +08:00
  • a5764116c5 Merge remote-tracking branch 'upstream/main' into push-to-pr-2055 yyj 2026-06-25 20:37:26 +08:00
  • a0c7431187
    ci(release-pypi): make release pipeline self-consistent (#2062) Benjamin 2026-06-25 20:05:06 +08:00
  • 59f028ff80 [build]: sync sglang submodule to 5d6bef9f61637aaeaf047bf8209def2af3eaa83f ovowei 2026-06-25 09:11:05 +00:00
  • 42fb35eb7a ci(release-pypi): make release pipeline self-consistent and Blackwell-ready yyj 2026-06-25 15:24:03 +08:00
  • 808b549c16 deploy: 983a88b620 yyj6666667 2026-06-24 07:23:06 +00:00
  • 7c752ebd30 deploy: 08d70e8605 yyj6666667 2026-06-24 07:22:49 +00:00
  • ecdeeb9923 deploy: 983a88b620 yyj6666667 2026-06-24 07:14:42 +00:00
  • 983a88b620
    docs: add FAQ entry about SGLang MoE kernel config warning (#2029) Willow Lopez 2026-06-24 15:12:54 +08:00
  • 8db3561516
    Update GPU architecture support information Benjamin 2026-06-24 15:12:21 +08:00
  • 08d70e8605
    [fix](kt-kernel): enable -mavx512vl for AVX512_VBMI/BF16 multi-variant builds (#2021) Hermann Hans Klie 2026-06-24 09:05:36 +02:00
  • 7dffc947e5
    Clean up AVX512 flags section in CMakeLists.txt Benjamin 2026-06-24 15:04:56 +08:00
  • 0a7785423a [fix](security): HMAC-authenticate scheduler RPC to prevent pickle deserialization RCE devangpratap 2026-06-23 17:55:56 -04:00
  • 9924c9f1bb deploy: 1ed332b6ea jdai0 2026-06-23 12:52:37 +00:00
  • 1ed332b6ea
    [docs]: clean up Qwen3.5 KT LoRA serving guide (#2057) Jiaheng Dai 2026-06-23 20:52:23 +08:00
  • 4e10cb4a7c
    Update doc/zh/Qwen3.5-SGLang-LoRA-Serving_zh.md Jiaheng Dai 2026-06-23 20:44:52 +08:00
  • d46d5b65b8 [docs]: clean up Qwen3.5 KT LoRA serving guide djh 2026-06-23 12:40:47 +00:00
  • 19fc69d597 deploy: c884dbd221 yyj6666667 2026-06-21 09:55:38 +00:00
  • c884dbd221
    release: bump version to 0.6.3 (#2054) Benjamin 2026-06-21 17:55:24 +08:00
  • e03c014465 release: bump version to 0.6.3 yyj 2026-06-21 17:51:25 +08:00
  • ce7c3ddbe9
    [build]: sync sglang submodule to 8b636f9008dbad58c0a8e481b03e794739e6c146 (#2047) v0.6.3 github-actions[bot] 2026-06-21 17:37:52 +08:00
  • 597e051a6d [build]: sync sglang submodule to 8b636f9008dbad58c0a8e481b03e794739e6c146 ovowei 2026-06-21 09:19:09 +00:00
  • 6c9c95601d
    [fix](kt-kernel): CudaGraph replay fix and Add MoE startup log (#2037) Benjamin 2026-06-18 17:15:37 +08:00
  • 3ffa8871aa fix(kt-kernel/cpuinfer): drop delete args — UAF under CUDA Graph replay yyj 2026-06-18 16:41:30 +08:00
  • f95dd3f0d2 [fix](kt-kernel): clarify AMX vs AVX512-BF16 backend in MoE startup log yyj 2026-06-09 13:19:45 +08:00
  • 73d611fe2a revert(kt-kernel/cpuinfer): drop delete args from sync_ callback yyj 2026-06-18 16:32:20 +08:00
  • 943cc4daeb
    [feat] MXFP8 MoE support (#2041) Benjamin 2026-06-18 16:18:49 +08:00
  • a50a505408 refactor(avx2/mxfp8): wire backend + swigluoai activation yyj 2026-06-18 15:21:00 +08:00
  • 21d5b40323
    docs(deepseek-v4-flash): pin apache-tvm-ffi<0.1.12 to avoid tilelang TVM FFI registration clash (#2045) (#2048) Tai An 2026-06-17 20:10:56 -07:00
  • 50d9434800
    [docs](kt-kernel): add MiniMax-M3 SGLang + KT-Kernel tutorial (#2051) Benjamin 2026-06-18 10:40:51 +08:00
  • 3b02827282 [docs](kt-kernel): add MiniMax-M3 SGLang + KT-Kernel tutorial yyj 2026-06-18 10:36:47 +08:00
  • ee0274e63b docs(deepseek-v4-flash): move apache-tvm-ffi pin rationale into a Note blockquote Tai An 2026-06-17 15:19:17 -07:00
  • 03c5448dd8
    Merge pull request #1 from yyj6666667/refactor/minimax-m3 Benjamin 2026-06-17 14:49:16 +08:00
  • cc3d9a65bd deploy: 0f2e7905e2 ovowei 2026-06-17 06:26:21 +00:00
  • 0f2e7905e2
    Update README.md (#2049) Jianwei Dong 2026-06-17 14:24:23 +08:00
  • a4fef71aff
    Update README.md ovowei-patch-3 Jianwei Dong 2026-06-17 14:24:11 +08:00
  • 0b4b5145d2 docs(deepseek-v4-flash): pin apache-tvm-ffi<0.1.12 to avoid tilelang TVM FFI clash Tai An 2026-06-16 18:16:15 -07:00
  • 4532ebce18 [fix](mxfp8): strict group_size==32 + dim divisibility in derived_init yyj 2026-06-16 23:11:23 +08:00
  • 0e2ff783d8
    update glm52 tutorial (#2046) Jianwei Dong 2026-06-16 17:15:23 +08:00
  • 542d146e51 glm52 tutorial support-glm52-djw ovowei 2026-06-16 17:03:08 +08:00
  • 7641f5445d
    [feat] : Add MACA backend support for kt-kernel and fix CPU MoE tests (#2044) Dayuxiaoshui 2026-06-16 16:49:06 +08:00
  • 39f2fc6b9d fix(kt-kernel): delete SyncArgs allocated for cudaLaunchHostFunc callback yyj6666667 2026-06-16 16:46:36 +08:00
  • 2dc584084c Fix AMX build flags and CPU MoE tests Dayuxiaoshui 2026-06-16 02:50:09 +00:00
  • 93b0d19b1b Add MACA event API mappings Dayuxiaoshui 2026-06-16 01:39:42 +00:00
  • 48b628c358 fix(balance_serve): bind scheduler RPC to loopback to close pre-auth pickle RCE sysy 2026-06-15 21:20:05 -04:00
  • 54e471796e Add MACA backend support for kt-kernel Dayuxiaoshui 2026-06-15 10:21:36 +00:00
  • 66ca4ce22d [fix](amx/mxfp8): unified per-row scatter + uint8 ue8m0 scale write yyj 2026-06-13 16:39:21 +08:00
  • 73ddff5c31 fix(avx2/mxfp8): unified per-row scatter for write_weights_to_buffer yyj 2026-06-12 16:20:56 +08:00
  • 9496a056a6 fix: add language_model.model prefix for M3 VL expert weight loading yyj 2026-06-06 15:25:12 +08:00
  • 1110ef9b0e fix: plumb swiglu_alpha through _create_inference_wrapper for MXFP8 yyj 2026-06-06 14:57:23 +08:00
  • bd79ad8840 [feat](kt-kernel): add MXFP8 Python integration (method, loader, wrapper) yyj 2026-06-02 23:07:32 +08:00
  • bd71ad3dc7 [feat](kt-kernel): add MXFP8 MoE kernel for MiniMax M3 Preview yyj 2026-06-02 22:50:13 +08:00
  • 89d30a3d01
    [fix(loader)]: correct off-by-one expert-count guard in load_experts (#2026) devangpratap 2026-06-07 11:41:04 -04:00
  • a8ac75ff36 chore: improve ktransformers maintenance path Noa Levi 2026-06-05 11:36:48 +00:00
  • c1cb22311b
    [build]: sync sglang submodule to 51032b71279d9038058563f8d2e758d99b278ef4 (#2032) github-actions[bot] 2026-06-05 19:24:22 +08:00
  • 60eb08f8d8 [build]: sync sglang submodule to 51032b71279d9038058563f8d2e758d99b278ef4 ovowei 2026-06-05 09:13:18 +00:00
  • a14b6dd5ec deploy: c9a915e6ac jdai0 2026-06-05 08:57:40 +00:00
  • c9a915e6ac
    [feat](kt-lora): add end-to-end Qwen3.5 MoE KT LoRA serving workflow (#2031) Jiaheng Dai 2026-06-05 16:57:14 +08:00
  • 84837552fa [fix](kt-lora): merge official main djh 2026-06-03 02:54:30 +00:00
  • ed11bcf5df [fix](kt-lora): validate adapter conversion djh 2026-06-02 13:12:18 +00:00
  • 14191ff407 docs: add trailing newline to FAQ.md Willow Lopez 2026-06-02 02:00:02 +08:00
  • 220f16b9de [feat](kt-lora): add merged SGLang adapter workflow djh 2026-06-01 12:06:27 +00:00
  • 762c6ebd5c
    Apply suggestion from @gemini-code-assist[bot] Willow Lopez 2026-05-31 20:57:14 +08:00
  • 1f6ca08742 docs: add FAQ entry about SGLang MoE kernel config warning Autumn 2026-05-31 13:47:36 +08:00
  • 97c033d817 [fix(docker)]: build local kt-kernel in fine-tune env before installing ktransformers devangpratap 2026-05-30 20:23:56 -04:00
  • ccbc867ff1 [test(loader)]: import loader as a top-level module in expert-count guard test devangpratap 2026-05-30 20:22:29 -04:00
  • e2aa18d8b8 [fix(docker)]: build fine-tune env from source instead of the unpublished release wheel devangpratap 2026-05-30 20:04:13 -04:00
  • 6d80523f7c [fix(loader)]: correct off-by-one expert-count guard in SafeTensorLoader.load_experts devangpratap 2026-05-30 18:46:58 -04:00
  • 870a628f2b
    Merge 25fa2fd532 into d41f569e84 Pory 2026-05-30 20:18:00 +00:00
  • d41f569e84
    [fix](cli): detect SGLANG_DSV4_2604_SUBMODE conflict before launch (#2025) devangpratap 2026-05-30 07:20:47 -04:00
  • ef6c47f9d2
    [feat](kt-kernel): AVX2 MXFP4 MoE MXFP4 dispatch (#2015) Benjamin 2026-05-30 19:20:16 +08:00
  • 97c5dcac88 feat(mesh): add rolling prefetch + runtime helpers for paper benchmarks RaQiu 2026-05-30 18:45:23 +08:00
  • c81cd099d0 Remove unused AMX tile MXFP4 kernel (GemmKernel224MXFP4) yyj 2026-05-30 14:31:00 +08:00
  • 08b8ab001e [feat]: pin Qwen3.5 non-expert LoRA support djh 2026-05-27 08:39:31 +00:00
  • 230d476fa7 [fix](cli): tighten env-var validation per review feedback devangpratap 2026-05-27 03:07:41 -04:00
  • a0fff3c0d4 [fix](cli): detect SGLANG_DSV4_2604_SUBMODE conflict before launch devangpratap 2026-05-27 03:00:52 -04:00
  • 072cb350a3 [feat](kt-lora): add KT expert LoRA adapter serving djh 2026-05-26 07:46:08 +00:00