minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity
16,360 Commits 3 Branches 0 Tags
3c533acec6cdb0b0e44ab0f26411e499225fef02
Commit Graph
100 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
Xiaoyu ZhangandClaude Opus 4.8 441910f926 [diffusion] LTX-2 quality=high fused RMSNorm+modulate + FFN GELU epilogue (H200 ltx23-one-stage denoise 45.85->43.24 s, ~matches torch.compile) (#34172) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-08-10 09:46:17 +08:00
Xiaoyu ZhangandClaude Opus 4.8 56ef810cad [diffusion] BCG: auto-capture the default warmup resolution instead of hard-requiring --warmup-resolutions (H200 SANA denoise 0.73->0.457 s with a single flag) (#34174) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-08-10 08:37:32 +08:00
Xiaoyu Zhang f6cbdc1dd1 docs(diffusion): refresh skills for latest runtime (#34143) 2026-08-09 10:30:46 +08:00
Xiaoyu Zhang 38c007dfe5 [diffusion] FLUX.1: route the adaLN LN+modulate sites through the bit-exact fused LayerNorm+modulate kernel (H200 1024^2 lossless denoise -1.2%, e2e wall -2.9%) (#34126) 2026-08-09 09:52:37 +08:00
Xiaoyu ZhangandClaude Fable 5 6424fec326 [diffusion] Bit-exact data-movement elimination for the Wan causal VAE decoder (H200 LongLive2 704x1280x61f: decode 2.80->2.32 s lossless / 2.12->1.67 s quality=high, e2e -10.7%) (#34125) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-09 09:50:56 +08:00
Xiaoyu ZhangandClaude Fable 5 33ed5d4413 [diffusion] perf_logger: SYNC_STAGE_PROFILING must drain the GPU queue for stage records too (fixes 2-3x inflated DecodingStage readings) (#34124) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-09 09:49:52 +08:00
Xiaoyu Zhang dc9624deb2 [diffusion] Clean up kernels and shared fast paths (#34085) 2026-08-09 00:37:00 +08:00
Xiaoyu ZhangandClaude Fable 5 891445676c [diffusion] Sana: bit-exact fused aten LayerNorm+modulate under BCG (H200 denoise -4.8%) (#34015) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-08 16:05:25 +08:00
Xiaoyu ZhangandClaude Fable 5 5dffa06fe1 [diffusion] GLM-Image bit-exact fused aten LayerNorm+modulate / qk-LN (H200 30-step denoise -8.1%) (#34008) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-08 13:26:11 +08:00
Xiaoyu Zhang 148f15b0af [diffusion] FLUX.1 fused adaLN modulate (bit-exact) + RoPE cache hoist, LN-affine folding behind quality=high (H200 e2e -3.5% lossless / -6.9% high) (#34004) 2026-08-08 13:07:42 +08:00
Xiaoyu ZhangandClaude Fable 5 6c7498113f [diffusion] Enable breakable CUDA graph for SANA (H200 1024px e2e -26%, bit-exact) (#33989) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-07 23:54:39 +08:00
Xiaoyu Zhang d4be483efb [diffusion] Enable breakable CUDA graph for LTX-2 (H200 two-stage e2e 10.75 s -> 6.90 s, 1.56x) (#33885) 2026-08-07 22:29:14 +08:00
Xiaoyu ZhangandClaude Fable 5 572434e2f6 [diffusion] Z-Image bit-exact fused qk-norm (H200 Turbo 1024px e2e -6.4%) (#33886) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-07 17:03:28 +08:00
Xiaoyu ZhangandClaude Fable 5 591cfb0881 [diffusion] FLUX.2 bit-exact residual-gate fast path (H200 klein-4B 50-step denoise -1.2%) (#33823) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-06 22:54:40 +08:00
Xiaoyu ZhangandClaude Fable 5 dd98c9572a [diffusion] Generalize the FLUX.2 VAE decoder fast path to AutoencoderKL (Z-Image / FLUX.1) behind quality=high (#33818) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-06 22:53:15 +08:00
Xiaoyu ZhangandClaude Fable 5 3654740347 [diffusion] ERNIE-Image bit-exact fused RMSNorm+scale/shift (H200 1024^2 e2e 15.63 -> 15.00 s, denoise -3.3%) (#33854) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-06 19:58:44 +08:00
Xiaoyu ZhangandClaude Fable 5 295784723a [diffusion] Ideogram 4: fuse RMSNorm modulate/gate chains via the Z-Image Triton suite behind quality=high (H200 e2e -2.9%/-3.4%) (#33822) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-06 19:57:34 +08:00
Xiaoyu ZhangandClaude Fable 5 eff6a11350 [diffusion] FLUX.1 bit-exact residual-gate fast path + tanh-GELU epilogue behind quality=high (H200 e2e -1.1% lossless / -4.3% high) (#33819) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-06 19:56:38 +08:00
Xiaoyu ZhangandClaude Fable 5 b6876fc652 [diffusion] ERNIE-Image bit-exact residual-gate fast path (H200 1024^2 e2e 16.17 -> 15.75 s) (#33734) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-06 13:43:09 +08:00
Xiaoyu ZhangandClaude Fable 5 4c0a8940fa [Kernel] Unify BaseFusedOp and MultiPlatformOp dispatch (#33205) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-06 08:52:09 +08:00
Xiaoyu ZhangandMohammad Miadh Angkad ba12a16a62 [diffusion] Prefer cuDNN SDPA over FA4 for dense attention on sm_100 (B200) (#33655) 
Co-authored-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com>
 2026-08-06 08:49:10 +08:00
Xiaoyu ZhangandClaude Fable 5 3425c93666 [diffusion] Wan VAE RMSNorm+SiLU fusion behind quality=high (H200 FastWan2.2 e2e 9.611 -> 9.125 s) (#33546) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-05 21:33:35 +08:00
Xiaoyu ZhangandClaude Fable 5 a5888c956f [diffusion] Pack Ulysses Q/K/V input all-to-all into one collective + reusable a2a staging buffers (#33667) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-05 19:15:46 +08:00
Xiaoyu ZhangandClaude Fable 5 95d0e57e83 [diffusion] Fuse DiT FFN tanh-GELU into up-proj GEMM (cublasLt epilogue) behind quality=high (Qwen-Image 1024^2 denoise 12.36 -> 12.05 s on H200) (#33536) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-04 23:49:43 +08:00
Xiaoyu Zhang 0d0c7d853f [diffusion] FLUX.2 VAE decoder fast path behind quality=high (H200: 1024^2 97.6->29.2 ms, 2048^2 437.2->168.5 ms) (#33451) 2026-08-04 23:48:19 +08:00
Xiaoyu ZhangandClaude Fable 5 c6f2a9c1d4 [diffusion] Restrict request-level quality to two validated tiers: lossless (default) and high (#33453) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-04 11:43:16 +08:00
Xiaoyu Zhang f829fb3ff7 [diffusion] Fix component accuracy topology reuse (#33317) 2026-08-04 08:35:03 +08:00
Xiaoyu ZhangandClaude Fable 5 4ef1660cd8 [Docs] MiniMax-H3: add measured H200 Ulysses4 vs TP2+Ulysses2 topology data (#33398) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-03 12:54:29 -07:00
Xiaoyu Zhang b64fd800d4 docs(diffusion): update skills for MiniMax-H3 (#33282) 2026-08-03 12:44:01 +08:00
Xiaoyu ZhangClaude Opus 5hnyls2002zhangxiaolei
 		fb207b72b0 feat(kernels): port standalone Kimi K3 kernels (#32890) 
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: hnyls2002 <lsyincs@gmail.com>
Co-authored-by: zhangxiaolei <zhangxiaolei.666@bytedance.com>
 2026-08-01 13:26:47 +08:00
Xiaoyu Zhang c5bd3d7dce [diffusion][benchmark] Add reproducible request-manifest offline benchmark (#32917) 2026-07-30 22:11:18 +08:00
Xiaoyu Zhang 7784ac8f91 [diffusion][docs] Fix Cosmos3 model sizes (#32916) 2026-07-30 22:10:23 +08:00
Xiaoyu Zhang 2e9c82b359 [Kernel] Remove unreachable AOT headers (#32842) 2026-07-30 22:08:57 +08:00
Xiaoyu Zhang 1d9c292547 [Kernel] Add inventory guards and clean benchmark layout (#32788) 2026-07-30 09:03:24 +08:00
Xiaoyu Zhang 0ebbe43dbb fix(diffusion): size VSA top-k from padded blocks (#32695) 2026-07-29 21:58:41 +08:00
Xiaoyu Zhang 4f5b50c576 perf(diffusion): decode Wan VAE in BF16 (#32697) 2026-07-29 21:57:50 +08:00
Xiaoyu Zhang 917e900d4d feat(diffusion): add regional torch compile (#32696) 2026-07-29 21:57:05 +08:00
Xiaoyu ZhangandClaude Fable 5 8742a1a0f8 Add a benchmark script for the HPC-Ops bf16xfp32 router GEMM (#32642) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-29 18:17:08 +08:00
Xiaoyu Zhang c32c4ef79c [Kernel] Move sgl-kernel under sglang.kernels.aot (#32648) 2026-07-29 17:25:00 +08:00
Xiaoyu Zhang c9947b087b Enable multimodal prefill BCG for VL and audio models (#30872) 2026-07-29 06:47:40 +08:00
Xiaoyu Zhang 7778dd23ea [diffusion] refactor: remove stale kernels and dead code (#32651) 2026-07-29 06:23:23 +08:00
Xiaoyu Zhang 9cffc2ba52 [Kernel] Remove unused implementations and stale registry entries (#32636) 2026-07-28 18:12:10 +08:00
Xiaoyu ZhangClaude Fable 5Halcyon
 		8d6549bc40 [Attention Backend] Extend hpc_ops dynamic-scheduled decode to bf16 (#32304) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Halcyon <56064364+VAthree@users.noreply.github.com>
 2026-07-27 21:31:04 +08:00
Xiaoyu ZhangClaude Fable 5Halcyon
 		3d91a569ce [MoE Backend] Add HPC-Ops FP8 MoE runner backend (#30541) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Halcyon <56064364+VAthree@users.noreply.github.com>
 2026-07-24 19:46:11 +08:00
Xiaoyu ZhangHalcyonClaude Fable 5
 		841fa293b5 [Fix] Reject online weight updates while the HPC-Ops router GEMM split cache is active (#31943) 
Co-authored-by: Halcyon <56064364+VAthree@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-24 19:30:13 +08:00
Xiaoyu ZhangandClaude Fable 5 d4a0dfbc31 [Fix] Two root causes of the H100 deepep TBO CI break: scale-tensor use-after-free + missing non-finite quant sanitization (#32188) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-24 07:29:28 +08:00
Xiaoyu ZhangandClaude Opus 4.8 62aa85d9aa [Kernel] Sweep missed dedicated kernels into kernels.ops (moe/quant siblings + dspark) (RFC #29630) (#32160) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-07-23 17:07:16 +08:00
Xiaoyu ZhangandClaude Opus 4.8 11b0e5c5ad [Kernel] Classification cleanup: unify _jit_ naming, drop empty/model groups, add elementwise (RFC #29630) (#32148) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-07-23 13:47:02 +08:00
Xiaoyu ZhangandClaude Opus 4.8 2d1a7be8c4 [Kernel] Reclassify kernel tests by ops group + move helpers out of the package (RFC #29630) (#32128) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-07-23 12:18:27 +08:00
Xiaoyu ZhangandClaude Opus 4.8 99f636a86f [Kernel] RFC #29630 finale: retire sglang.jit_kernel into sglang.kernels (#32072) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-07-23 08:35:09 +08:00
Xiaoyu ZhangClaude Fable 5Halcyon
 		0a6d1930c3 [Attention Backend] Add HPC-Ops attention backend (#30540) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Halcyon <56064364+VAthree@users.noreply.github.com>
 2026-07-22 22:06:22 +08:00
Xiaoyu ZhangandClaude Opus 4.8 74338e94f1 [Kernel] Phase 4 batch-3: migrate tangled JIT subsystems + new groups into kernels.ops (RFC #29630) (#32045) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-07-22 21:15:03 +08:00
Xiaoyu ZhangandClaude Opus 4.8 977ea336cd [Kernel] Phase 4 batch-2: migrate JIT operator groups into kernels.ops (no shims) (RFC #29630) (#32015) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-07-22 17:49:51 +08:00
Xiaoyu ZhangandClaude Opus 4.8 246b3c3eaf [Kernel] Phase 3+4: move JIT infra + operator groups into sglang.kernels (RFC #29630) (#31666) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-07-22 11:28:27 +08:00
Xiaoyu Zhang 075bd97952 [Benchmark] Remove obsolete auto-benchmark remnants (#31941) 2026-07-21 20:44:52 +08:00
Xiaoyu ZhangClaude Fable 5Halcyon
 		e4eea7ce2f Optimize LongCat-Flash router GEMM with the HPC-Ops bf16xfp32 kernel (#30247) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Halcyon <56064364+VAthree@users.noreply.github.com>
 2026-07-21 20:05:17 +08:00
Xiaoyu Zhang 829e9ce9d5 Lower AutoRound quantization MMLU threshold (#31748) 2026-07-20 13:40:48 +08:00
Xiaoyu ZhangandClaude Opus 4.8 216b750c8f [Kernel] Sweep decoupled scattered kernels into sglang.kernels.ops (RFC #29630) (#31582) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-07-18 19:07:07 +08:00
Xiaoyu ZhangandClaude Opus 4.8 619609aa5a [Kernel] Simplify sglang.kernels tests to idiomatic pytest style (RFC #29630) (#31546) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-07-17 15:06:26 +08:00
Xiaoyu ZhangandClaude Opus 4.8 1ac1ffea0c [Kernel] Fill non-CUDA coverage: HIP (aiter/rocm-triton) + Ascend NPU backends (RFC #29630) (#31307) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-07-17 14:05:35 +08:00
Xiaoyu ZhangandClaude Opus 4.8 8432eafd3d [Kernel] Decouple KernelBackend from device + device-based CapabilityRequirement (RFC #29630) (#31292) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-07-17 10:35:34 +08:00
Xiaoyu Zhang e73f323464 [JIT] Reduce MoE fused gate CI test sweep (#31400) 2026-07-16 14:42:45 +08:00
Xiaoyu ZhangandClaude Fable 5 22453ca63c docker: build HPC-Ops into the GPU image (#31390) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-16 11:07:01 +08:00
Xiaoyu ZhangandClaude Fable 5 4aadf94146 [Kernel] Relocate vendored fla and mamba kernel trees to sglang.kernels (RFC #29630, Phase 2.5, 7/7) (#30795) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-15 12:52:15 +08:00
Xiaoyu ZhangandClaude Fable 5 c00131ebaa [Kernel] Migrate linear-attention, MiniMax-sparse and diffusion kernels to sglang.kernels (RFC #29630, Phase 2.5, 6/7) (#30793) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-15 11:21:36 +08:00
Xiaoyu ZhangandClaude Fable 5 ba5be86d42 [Kernel] Migrate DSA + DSV4 attention kernels to sglang.kernels (RFC #29630, Phase 2.5, 5/7) (#30792) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-15 11:11:22 +08:00
Xiaoyu ZhangandClaude Fable 5 ee000f6734 [CI] Fix SGLANG_JIT_KERNEL_RUN_FULL_TESTS never activating the nightly full jit-kernel sweep (#31042) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-14 17:32:29 +08:00
Xiaoyu ZhangandClaude Fable 5 1a35440c4a [Kernel] Migrate generic attention kernels to sglang.kernels (RFC #29630, Phase 2.5, 4/7) (#30789) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-14 16:53:46 +08:00
Xiaoyu ZhangandClaude Fable 5 e9ef06c560 [Kernel] Migrate top-level srt/layers stray kernels to sglang.kernels (RFC #29630, Phase 2.5, 3/7) (#30787) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-14 09:20:59 +08:00
Xiaoyu ZhangandClaude Fable 5 ee464fedc6 [Kernel] Migrate scattered MoE kernels to sglang.kernels (RFC #29630, Phase 2.5, 2/7) (#30786) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-14 09:03:21 +08:00
Xiaoyu ZhangandClaude Opus 4.8 4c997310f5 [Kernel] Hotfix: update sgl-kernel imports of relocated fp8_kernel (RFC #29630 #30784) (#31089) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-07-14 08:41:23 +08:00
Xiaoyu ZhangandClaude Fable 5 874fc07d9b [Kernel] Migrate scattered quantization kernels to sglang.kernels (RFC #29630, Phase 2.5, 1/7) (#30784) 
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-07-13 16:17:01 +08:00
Xiaoyu Zhang 65abb23842 Add diffusion BCG prompt conditioning guard (#30782) 2026-07-11 13:14:31 +08:00
Xiaoyu ZhangandClaude Opus 4.8 6ed9843b57 [Kernel] Introduce sglang.kernels namespace and migrate scattered triton_ops kernels (RFC #29630, Phase 2) (#30044) 
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-07-10 21:41:08 +08:00
Xiaoyu Zhang e9493a015c Fix diffusion BCG lifetime and add Z-Image-Turbo CI (#30584) 2026-07-10 21:19:59 +08:00
Xiaoyu Zhang b8ca06fdad Fix zero expert routed ids for MoE backends (#30387) 2026-07-08 21:23:25 +08:00
Xiaoyu ZhangBBufClaude Opus 4.8BBuf
 		33c3dfd7e0 [diffusion] Enable breakable CUDA graph (BCG) for diffusion DiTs (#27436) 
Co-authored-by: BBuf <xiaoyu.zhang@radixark.net>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: BBuf <bbuf@sglang.local>
 2026-07-08 14:45:48 +08:00
Xiaoyu ZhangandZijie Xia ead1e490b5 [Doc] Add LongCat 2.0 FP8 cookbook (#30320) 
Co-authored-by: Zijie Xia <zijie.xia@radixark.ai>
 2026-07-07 11:48:13 -07:00
Xiaoyu Zhangsunjiaqi11Ke Bao
 		e339c83f82 [Model] Support LongCat 2.0 FP8 (#30275) 
Co-authored-by: sunjiaqi11 <sunjiaqi11@meituan.com>
Co-authored-by: Ke Bao <ispobaoke@gmail.com>
 2026-07-07 19:51:12 +08:00
Xiaoyu ZhangandClaude Opus 4.8 1da7d3a50b [MoE] Retire the AOT moe_fused_gate / kimi_k2_moe_fused_gate gate kernels (#26771) (#29997) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-07-07 13:53:17 +08:00
Xiaoyu Zhang 931b00f1b0 [diffusion] Clean up duplicate helper definitions (#30159) 2026-07-05 22:05:11 +08:00
Xiaoyu Zhang 6dd0cefb2a [CI] Revert ModelOpt NVFP4 threshold relax (#29844) 2026-07-04 21:17:32 +08:00
Xiaoyu ZhangandClaude Opus 4.8 a2d7eb303e [MoE] Consolidate ungrouped + grouped gate/topk onto one Triton router (#26771) — faster than AOT on B200/H100/H200, at parity with flashinfer (#29771) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-07-03 11:18:59 +08:00
Xiaoyu Zhang b276a9acee chore: cleanup garbage code (#29770) 2026-07-02 16:14:01 +08:00
Xiaoyu ZhangKhoa PhamClaude Opus 4.8Mohammad Miadh Angkad
 		df0dfbaa45 [Kernel] Strengthen kernel shape coverage (#29636) 
Co-authored-by: Khoa Pham <khoa.pham@radixark.ai>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com>
 2026-07-01 15:44:20 +08:00
Xiaoyu Zhang 8205aa3603 chore: clean diffusion dead code (#29789) 2026-07-01 15:42:16 +08:00
Xiaoyu Zhang 47ae1241d3 [CI] Relax ModelOpt NVFP4 diffusion consistency thresholds (#29767) 2026-07-01 14:45:08 +08:00
Xiaoyu Zhang fcb9f229b3 [KDA-Pilot] Add LTX2 QKNorm split-RoPE CUDA fast path (#29708) 2026-07-01 14:42:07 +08:00
Xiaoyu ZhangandClaude Opus 4.8 3add35e26d [Diffusion] Reuse shared AlignedVector and tidy jit_kernel/diffusion (#29664) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-06-30 11:38:22 +08:00
Xiaoyu Zhang c36f166364 [skill] Remove outdated llm-serving-auto-benchmark skill (#29487) 2026-06-27 14:19:11 +08:00
Xiaoyu ZhangandClaude Opus 4.8 495f13fa12 [KDA-Pilot] Add diffusion residual-gate CUDA fast path for LTX2 (#29361) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-06-27 12:59:41 +08:00
Xiaoyu Zhang 8524678889 Fix MiniMax MSA fallback when fmha plan is unavailable (#29250) 2026-06-26 23:14:31 +08:00
Xiaoyu Zhang 18b0e5757e [Diffusion] Fuse LTX2 Ada values (#29390) 2026-06-26 23:13:15 +08:00
Xiaoyu ZhangandClaude Opus 4.8 5996b54bd3 [KDA-Pilot] Add diffusion causal Conv3D cat-pad CUDA fast path for Cosmos3 (#29281) 
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
 2026-06-26 15:06:49 +08:00
Xiaoyu Zhang 4d06d4c97f Sync Gemma4 hardware table with Blackwell recipes (#29266) 2026-06-25 22:57:33 +08:00
Xiaoyu Zhangjingyu-mlClaude Opus 4.8
 		52c32035eb [diffusion] Add Qwen-Image ModelOpt NVFP4 support (#28928) 
Co-authored-by: jingyu-ml <jingyux@nvidia.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
 2026-06-25 22:56:09 +08:00
Xiaoyu Zhang ddda4f9028 [Bugfix] Fix Ministral3 init argument forwarding (#29111) 2026-06-25 18:26:45 +08:00
Xiaoyu Zhang 7c9804ef21 Add MiMo V2.5 Blackwell vision FA4 recipe (#29253) 2026-06-25 13:47:32 +08:00
Xiaoyu Zhang efbe67d237 Tune Gemma4 26B-A4B B200 memory recipe (#29252) 2026-06-25 11:31:25 +08:00
Xiaoyu Zhang 26e1d4d847 [KDA-Pilot] Add B200 diffusion norm-scale-shift CUDA fast path for Qwen-Image (#27392) 2026-06-24 14:38:36 +08:00