 Xiaoyu Zhang
|
be00a543a7
|
perf: use Gumbel-max trick in the main sampler to cut decode CPU dispatch (#38117)
|
2026-09-06 22:40:27 +08:00 |
|
|
MickandClaude Fable 5.1
|
a176ba2f7b
|
[diffusion] feat: measure warmup memory and layer usage per phase for residency calibration (1/4) (#37916)
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
|
2026-09-06 19:42:42 +08:00 |
|
|
Mick
|
938dc5621d
|
[diffusion] refactor: reuse plain state-dict loading without per-model classes (#38127)
|
2026-09-06 18:39:21 +08:00 |
|
|
+2
|
97c6978369
|
GLM-5.3-Flash support (#36507)
Co-authored-by: zRzRzRzRzRzRzR <Yuxuan.Zhang2@liverpool.ac.uk>
Co-authored-by: Shijin Zhang <75300765+Dovis01@users.noreply.github.com>
Co-authored-by: zanes-ops <zanes@nvidia.com>
Co-authored-by: Baizhou Zhang <sobereddiezhang@gmail.com>
Co-authored-by: Jian Chen <jianchen0311@gmail.com>
Co-authored-by: zijiexia <37504505+zijiexia@users.noreply.github.com>
Co-authored-by: andyluo7 <43718156+andyluo7@users.noreply.github.com>
Co-authored-by: Ehsan Akhgari <ehsan.akhgari@gmail.com>
Co-authored-by: kpham-sgl <khoa.pham@radixark.ai>
Co-authored-by: BBuf <1182563586@qq.com>
Co-authored-by: Raiden Makoto <81530826+Raiden-Makoto@users.noreply.github.com>
|
2026-09-06 02:27:59 -07:00 |
|
|
Mick
|
a9944aec01
|
[diffusion] CI: guard the allocated vram peak with reporting the reserved one (#38172)
|
2026-09-06 17:22:16 +08:00 |
|
|
Mickandmickqian
|
8ef646a5c6
|
fix(vlm): contain EPD request lifecycle failures (#36944)
Co-authored-by: mickqian <mickqian@users.noreply.github.com>
|
2026-09-06 16:05:10 +08:00 |
|
|
Vincent Gao
|
ae54ccb25d
|
[Router] Publish cache-aware load state (#38139)
|
2026-09-06 15:22:25 +08:00 |
|
|
Zhang, Jiejing
|
6cee9285a3
|
[ROCm] Make DSA indexer top-k exact with cooperative selection (#37591)
|
2026-09-05 23:55:56 -07:00 |
|
|
Mick
|
e3f7097591
|
[diffusion] refactor: consolidate plain state-dict component loaders (#38128)
|
2026-09-06 13:52:54 +08:00 |
|
|
Depend
|
67e3ccda97
|
[diffusion] fix: restore non-layer placeholders before releasing host copies (#38171)
|
2026-09-06 13:41:30 +08:00 |
|
|
Mick
|
febb360519
|
[VLM] retire aborted disaggregated prefill results (#36988)
|
2026-09-06 10:15:33 +08:00 |
|
|
Liangsheng Yin
|
f5819b09bf
|
Revert "[AMD][DSV4] Fix unified-KV pool sizing and SWA ring accounting" (#38163)
|
2026-09-05 17:28:46 -07:00 |
|
|
Mohammad Miadh AngkadandMohammad Angkad
|
09daea94ac
|
Support NoPE layers in the tokenspeed_mla FP8 prefill hook (#38152)
Co-authored-by: Mohammad Angkad <mohammad.angkad@radixark.ai>
|
2026-09-05 16:51:04 -07:00 |
|
|
yuttian1
|
514b45fd34
|
[AMD][DSV4] Fix unified-KV pool sizing and SWA ring accounting (#30315)
|
2026-09-05 16:39:40 -07:00 |
|
|
Alex NailsandClaude Opus 5
|
6a0c55fd6c
|
[CI] Pin the Rust TreeCore build to the resolved libtorch instead of interpreter discovery (#37696)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
2026-09-05 15:30:32 -07:00 |
|
|
|
77aee20259
|
[Model] Add support for Nanbeige4.2 (#32151)
Co-authored-by: root <lizongqiang@kanzhun.com>
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
|
2026-09-06 03:57:27 +08:00 |
|
|
Xiaoyu ZhangandWaterpine
|
ccf9fe6590
|
[Kernel] Add KDA FP8 skinny GEMM for SM120 (#38082)
Co-authored-by: Waterpine <biansonghz@gmail.com>
|
2026-09-05 22:27:06 +08:00 |
|
|
Xiaoyu Zhang
|
dc2843801d
|
perf(lfm2): fuse gating and short convolution on SM90 (#37622)
|
2026-09-05 21:52:16 +08:00 |
|
|
Beihao Zhou
|
1e6f18bfeb
|
[MoE Refactor] Migrate SM100 trtllm-gen mxfp4 MoE onto MoeRunner (#32405)
|
2026-09-05 13:48:10 +00:00 |
|
|
Xiaoyu Zhang
|
eda10c3678
|
[Diffusion] Enable breakable CUDA graph for JoyEcho (#38110)
|
2026-09-05 21:45:53 +08:00 |
|
|
Mick
|
5df60a21cd
|
fix(vlm): harden EPD receiver validation and liveness (#36945)
|
2026-09-05 21:22:37 +08:00 |
|
|
Mick
|
a18106bbc3
|
fix(vlm): make EPD cache publication transactional (#36949)
|
2026-09-05 20:33:23 +08:00 |
|
|
|
bd16c22a04
|
[diffusion] fuse LingBot MoE group-limited top-k index selection (#38044)
Co-authored-by: BBuf <bbuf@users.noreply.github.com>
Co-authored-by: Mick Qian <mickqian@users.noreply.github.com>
|
2026-09-05 18:12:30 +08:00 |
|
|
DayuxiaoshuiandXiaoyu Zhang
|
50c1bf0db0
|
[Diffusion] Port the Wan VAE decoder fast paths to the Qwen-Image VAE (#38020)
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>
|
2026-09-05 18:02:33 +08:00 |
|
|
hhhh1252023
|
0948e6ebed
|
[CI] Remove metrics artifact mechanism from nightly NPU workflows (#35489)
|
2026-09-05 17:16:52 +08:00 |
|
|
Xiaoyu ZhangandBBuf
|
d49180019b
|
fix(moe): cast filtered-activation expert_ids to int32 for torch.compile (#38085)
Co-authored-by: BBuf <bbuf@users.noreply.github.com>
|
2026-09-05 17:13:07 +08:00 |
|
|
Xiaoyu Zhang
|
a74470e904
|
fix(mamba): unify causal_conv1d col* dtype to x (MiniCPM-V-4.6 GDN prefill bf16/fp16 mismatch) (#38039)
|
2026-09-05 17:08:20 +08:00 |
|
|
jacky.cheng
|
0bdc15d20f
|
[AMD] Fix ROCm VAE Conv2D fast path breaking spatial-parallel decode (#34424)
|
2026-09-05 01:00:25 -07:00 |
|
|
Xiaoyu Zhang
|
da76fa073f
|
[diffusion] fix: fix host-resident vocab tables loaded on GPU (#38012)
|
2026-09-05 14:03:21 +08:00 |
|
|
Mick
|
0ea8378085
|
[diffusion] feat: support request-scoped skip-softmax attention (#37959)
|
2026-09-05 13:50:17 +08:00 |
|
|
Shuwen WangandClaude Opus 5
|
4b44a1cde2
|
[Refactor] Let eviction policies take construction parameters (#37795)
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
2026-09-05 13:48:30 +08:00 |
|
|
Xinyuan Tong
|
32a1d55431
|
fix(modelopt_fp4): skip NVFP4 swiglu-fusion interleave for shared experts with swiglu_limit (#37378)
|
2026-09-04 22:42:29 -07:00 |
|
|
Ke Bao
|
ae3205ba28
|
Fail fast on undersized swa pool (#37610)
|
2026-09-05 13:40:37 +08:00 |
|
|
Yash AkhauriandXiaoyu Zhang
|
756d0e0a85
|
[Model] Add K2 Horizon FP8 checkpoint support (#38033)
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>
|
2026-09-05 13:18:26 +08:00 |
|
|
|
e980c1a2f1
|
fix(glm4v): disambiguate mixed image video offsets (#37971)
Co-authored-by: duxin <xinheng.dx@alibaba-inc.com>
Co-authored-by: Xiaoyu Zhang <1182563586@qq.com>
|
2026-09-05 13:16:16 +08:00 |
|
|
Liangsheng Yin
|
0454c074b4
|
[mem_cache] Clean up unified allocator leftovers (#38103)
|
2026-09-04 22:02:50 -07:00 |
|
|
|
3a770da756
|
[Unified Tree] Support Branching-Point Caching for the SWA Component (#34565)
Co-authored-by: alphabetc1 <2508695655@qq.com>
Co-authored-by: Shuwen Wang <47200617+alphabetc1@users.noreply.github.com>
|
2026-09-05 12:29:16 +08:00 |
|
|
Po-Han Huang (NVIDIA)
|
991368d880
|
Fix Muse Glimmer ModelOpt mixed weight mapping (#37510)
|
2026-09-04 20:53:55 -07:00 |
|
|
|
92a4d8b5ee
|
Clean logging under --weight-loader-prefetch-checkpoints (#33930)
Co-authored-by: Brayden Zhong <brayden@radixark.ai>
Co-authored-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com>
|
2026-09-04 20:05:53 -07:00 |
|
|
Liangsheng Yin
|
0645398a32
|
[mem_cache] Move the unified-memory allocators into allocator/ and split the composites out (#38072)
|
2026-09-04 19:53:15 -07:00 |
|
|
Xiaoyu ZhangandBBuf
|
d6e0a8cbf4
|
[diffusion] add Helios per-token gated-residual fusion (quality-gated) (#38042)
Co-authored-by: BBuf <bbuf@users.noreply.github.com>
|
2026-09-05 10:29:09 +08:00 |
|
|
Niko MaandZhangheng
|
f1f2380d2b
|
[Unified Cache][6/N]: Add UMBP external linker (#37578)
Co-authored-by: Zhangheng <hzh0425@apache.org>
|
2026-09-05 09:21:43 +08:00 |
|
|
Xiaoyu ZhangandBBuf
|
bc727bc4ee
|
[FP8] SM120: route FP8 linear to per-tensor (cudnn/nvjet) instead of channelwise cutlass (#38006)
Co-authored-by: BBuf <bbuf@example.com>
|
2026-09-05 09:07:13 +08:00 |
|
|
|
55bf3380e0
|
Support Hy4-preview (#36805)
Co-authored-by: BBuf <1182563586@qq.com>
Co-authored-by: alphabetc1 <2508695655@qq.com>
|
2026-09-04 18:03:49 -07:00 |
|
|
Xiaoyu Zhang
|
85da5457de
|
[diffusion] auto-keep video DiT resident on high-memory GPUs (#38001)
|
2026-09-05 08:54:09 +08:00 |
|
|
|
3c2724c48d
|
[AMD] Optimize Kimi-K3 Triton MLA prefill on gfx950 (#35770)
Co-authored-by: clintg6 <7388379+clintg6@users.noreply.github.com>
Co-authored-by: Thomas Wang <thomawan@amd.com>
Co-authored-by: HAI <hixiao@gmail.com>
Co-authored-by: kk <43161300+kkHuang-amd@users.noreply.github.com>
|
2026-09-04 17:44:30 -07:00 |
|
|
Alex Nails
|
a44bb397a9
|
[Perf] Vectorize alloc_extend_naive to remove the per-request Python loop (#37938)
|
2026-09-04 17:26:25 -07:00 |
|
|
Lianmin Zheng
|
613d87becd
|
[Memory] Reuse output storage across full prefill CUDA graphs (#38038)
|
2026-09-04 17:23:31 -07:00 |
|
|
YAMY
|
65f7957142
|
fix: gather CP-sharded tokens before TP-sharded dense MLP under prefill CP (#38078)
|
2026-09-04 17:14:31 -07:00 |
|
|
Amy Chang
|
8a98f11078
|
[feature] Add response-level input/output token ids to chat completions via SglExt (#34488)
|
2026-09-04 16:45:13 -07:00 |
|