Commit Graph

6166 Commits

Author SHA1 Message Date
wxy
43eecd8265 [diffusion] feat: support efficient sequence shard (#18161) 2026-02-08 21:09:39 +08:00
Zack Yu
d71ccd8860 fix: sync server_args.kv_cache_dtype when detecting FP8 KV cache (#18394) 2026-02-08 14:10:59 +08:00
DarkSharpness
8e2e835c2f [Fix] Fix backend selection after flashinfer version update (#18364) 2026-02-08 11:20:41 +08:00
Makcum888e
00248d85c7 [diffusion] platform: support WAN/FLUX/Qwen-Image/Qwen-Image-edit on Ascend (#13662)
Co-authored-by: dhx98 <haox.dai@gmail.com>
Co-authored-by: DHX98 <haoxiand@andrew.cmu.edu>
Co-authored-by: ronnie_zheng <zl19940307@163.com>
Co-authored-by: DHX98 <DHX98@noreply.gitcode.com>
Co-authored-by: Yuhao Yang <47235274+yhyang201@users.noreply.github.com>
2026-02-08 10:45:30 +08:00
Mohammad Miadh Angkad
7b83659310 fix: fix NVFP4 Kimi-K2.5 weight mapping and exclude list (#18370) 2026-02-08 10:23:48 +08:00
wxy
64950d8f97 [diffusion] feat: support saving videos directly on the server to avoid the overhead of tensor transfer (#18253) 2026-02-07 22:08:42 +08:00
Mick
31d4cd2ffd [diffusion] fix: respect dist_timeout option (#18386) 2026-02-07 20:56:04 +08:00
Mohammad Miadh Angkad
fddef76619 [Doc] Fix outdated --fp4-gemm-backend documentation (#18350) 2026-02-07 20:42:47 +08:00
Hao Jin
d792aa7618 [diffusion] fix: remove unnecessary norm_type argument from GLM-Image dits (#18382)
Co-authored-by: Hao Jin <Hao Jin>
2026-02-07 20:35:12 +08:00
Baizhou Zhang
eb4cf1dfc4 [CI] Skip some flaky subtests for test_multi_lora_backend.py (#18408) 2026-02-07 19:06:53 +08:00
Xiaoyu Zhang
baec650462 [Diffusion] Apply fused_norm_scale_shift to LTX2/MOVA (#18257)
Co-authored-by: yihanc <yingluosanqian@gmail.com>
2026-02-07 17:28:42 +08:00
赵晨阳
1552aab741 Support execute_shell_command for env var support (#18390) 2026-02-07 12:33:29 +08:00
hlu1
4637970dfb [Qwen3Next] Optimize fused_sigmoid_gating_delta_rule_update_kernel (#18271) 2026-02-07 11:59:42 +08:00
Neal Vaidya
f1ff697494 add hybrid model PD to NIXL connector (#16229)
Signed-off-by: Neal Vaidya <nealv@nvidia.com>
Co-authored-by: ishandhanani <82981111+ishandhanani@users.noreply.github.com>
2026-02-06 15:05:36 -08:00
Prozac614
e13b727e92 [diffusion] CI: update perf baseline (#17512) 2026-02-07 00:28:44 +08:00
shaharmor98
c6aa1863be Add Nemotron 3 Nano tests (#18119)
Signed-off-by: Shahar Mor <smor@nvidia.com>
2026-02-06 23:55:42 +08:00
xiaoye
79d409f210 [diffusion] fix: offload text encoder model in image encoding stage (#18317) 2026-02-06 22:55:56 +08:00
Xuchun Shang
3d68bd9d9b add hicache jit test (#17847)
Signed-off-by: Xuchun Shang <xuchun.shang@linux.alibaba.com>
2026-02-06 16:54:33 +08:00
陈一涵
f798ab9775 [diffusion] fix: fix torch.compile graph break caused by torch._dynamo.disable (#18336) 2026-02-06 14:48:09 +08:00
Alison Shao
d0c39bc219 Fix cross-container HF download race condition in CI (#18328) 2026-02-05 21:01:41 -08:00
Linyu Wu
aa390d2762 [Kernel] Migrate GPTQ-Marlin GEMM kernel to JIT (#18067) 2026-02-06 08:31:42 +08:00
aaaandychen
6a4b81e2d9 Refactor(qwen3-vl) optimize position encoding interpolation (#16781)
Signed-off-by: chenzhenyang <andy271828@163.com>
Signed-off-by: chenzhenyang <chenzhenyang@moonshot.cn>
Co-authored-by: chenzhenyang <chenzhenyang@moonshot.cn>
Co-authored-by: zhaochenyang20 <zhaochen20@outlook.com>
2026-02-05 10:26:35 -08:00
ovidiusm
498d8d0680 NixlKVManager optimizations (#17654)
Signed-off-by: Ovidiu Mara <ovidium@nvidia.com>
2026-02-06 00:25:23 +08:00
wxy
b639779dd8 [diffusion] feat: allow T5's TP Group to reuse the transformer's SP Group (#17818)
Co-authored-by: Mick <mickjagger19@icloud.com>
2026-02-06 00:12:19 +08:00
Glen Liu
3f32a5831d throw error if got adapter with added_tokens (#18046) 2026-02-05 23:55:43 +08:00
pansicheng
2eb4359ada [Kernel] Add JIT apply_rope_with_cos_sin_cache_inplace (#18155) 2026-02-05 21:49:37 +08:00
陈一涵
4aa03d91fd [diffusion] fix: fix accuracy bug caused by #14717 (#18296) 2026-02-05 20:36:18 +08:00
Shangming Cai
afae4c7178 [PD] Minor code cleanup for mooncake backend (#18279)
Signed-off-by: Shangming Cai <csmthu@gmail.com>
2026-02-05 17:38:09 +08:00
zhangheng
079fc8f3c5 [piecewise graph]: support MiniMax-M2 (#18217) 2026-02-04 23:24:38 -08:00
danielafrimi
3f1df322f9 [FIX] Always support TP > 4 for FP4 Gemm (#17300) 2026-02-05 15:10:26 +08:00
Meng, Hengyu
368936a62b [XPU] Integrate MoE and minor improvements in XPU attention backend (#13561) 2026-02-04 23:09:59 -08:00
Xiaoyu Zhang
dff3ba202a [Diffusion] Support layerwise offload for mova (#18272) 2026-02-05 13:16:07 +08:00
Ch3ngY1
f730c18679 [PD] improve kv offset calculation for MHA model with different tp size (#18163)
Co-authored-by: Shangming Cai <csmthu@gmail.com>
2026-02-05 10:43:23 +08:00
Mick
f218234e4f [diffusion] chore: prohibit Chinese characters usage (#18249) 2026-02-05 09:22:26 +08:00
yinghui
599c5f4922 fix kimi k2.5's moe gemm config init (#18064) 2026-02-04 16:59:01 -08:00
linhaifeng
c1d5cc3b24 [Bugfix] fix a obvious logic error (#18254) 2026-02-04 13:59:58 -08:00
Mohammad Miadh Angkad
efbf39583e Add MoE fused config for Qwen3-Coder-Next-FP8 on H100 TP=2 (#18195) 2026-02-04 13:36:35 -08:00
Zack Yu
2e87c2bd5e fix: fix MockModelRunner in attention tests (#18240) 2026-02-04 13:18:02 -08:00
Michael
6fd878b41d [AMD] Add kimi mi35x nightly test, folder organization and several stability fixes (#17895) 2026-02-04 12:03:57 -08:00
Mick
36a3e78af9 [diffusion] refactor: move model_stages into stages folder (#18248) 2026-02-05 00:23:31 +08:00
RunningLeon
3e7ecb78a6 model: support interns1-pro (#18145)
Co-authored-by: Ke Bao <ispobaoke@gmail.com>
2026-02-05 00:22:44 +08:00
RunningLeon
a6f53cc5e3 entrypoint: support passing spaces_between_special_tokens per request (#17939) 2026-02-04 22:18:36 +08:00
wxy
4c403045ec [diffusion] fix: fix the bug of redundant memory usage on GPU-0 (#18221) 2026-02-04 21:25:23 +08:00
Zhang Yiyang (SII)
0c9a0adc53 [diffusion] chore: clean MOVA codes (#18107) 2026-02-04 21:23:41 +08:00
BingjiaWang
760ae933bb optimize get_topk_ragged by fusing get k and k_scale triton kernel (#16043)
Co-authored-by: abing <wangbingjia.wbj@alibaba-inc.com>
2026-02-04 19:59:41 +08:00
Nicolas Castet
315306d8a9 Make sure we always disable symm memory without dp padding (#18129) 2026-02-04 19:58:28 +08:00
Jincong Chen
a72f4f839c Tiny fix for fp8 moe backend flashinfer_trtllm naming (#18243) 2026-02-04 19:58:04 +08:00
Evrard-Nil
ce02df8592 [diffusion] logging: downgrade default prompt log from info to debug (#17813)
Co-authored-by: Mick <mickjagger19@icloud.com>
2026-02-04 19:19:02 +08:00
Cheng Wan
84c09913eb Moving _alloc_extend_naive out of npu allocator (#18200) 2026-02-04 02:09:55 -08:00
zhangheng
be557cbc5f [RadixTree][5/N Refactor]: Introduce pre and post-processing methods for key matching (#18147) 2026-02-04 17:10:46 +08:00