Commit Graph

5460 Commits

Author SHA1 Message Date
jiapingW
5d200dd8d9 [diffusion] bench: distinguish between video generation and image generation in the bench_serving (#16149) 2025-12-30 21:29:18 +08:00
Jumiar
664f611e83 Add profiling capture support to the encoder server (#15730)
Signed-off-by: liuanqi <liuanqi6@xiaomi.com>
Co-authored-by: liusy58 <liusy58@linux.alibaba.com>
2025-12-30 19:56:25 +08:00
yh0903
49adb37e37 [diffusion] chore: fix ZMQ binding and model loading for FastWan compatibility (#13978)
Co-authored-by: Han Yu <hyu5@dt-login01.delta.ncsa.illinois.edu>
Co-authored-by: Mick <mickjagger19@icloud.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
2025-12-30 18:09:38 +08:00
Gaoji Liu
7518dc3532 feat(SpecEagleV2): add standalone_worker_v2 (#12625)
Co-authored-by: liugaoji.lgj <liugaoji.lgj@alibaba-inc.com>
2025-12-30 17:55:04 +08:00
Yuan Luo
94bcc19bce [VLM] Support Video for InternVL3_5 (#15942)
Co-authored-by: luoyuan.luo <luoyuan.luo@antgroup.com>
2025-12-30 17:07:49 +08:00
Shangming Cai
db3821a9ef [PP] Add a minimum chunk value for PP dynamic chunking (#16140)
Signed-off-by: Shangming Cai <csmthu@gmail.com>
2025-12-30 16:35:39 +08:00
Liangsheng Yin
c2601f0d21 Fix wrong assigning extend_input_len_per_req with eagle. (#16129) 2025-12-30 15:51:53 +08:00
Vitaly Tuzov
1048803c1f Reworked fast_pos_embed_interpolate() using torch (#10959) 2025-12-30 14:45:34 +08:00
HuangJi
8a84b1e7e0 [diffusion] model: support TurboWan2.1-T2V-1.3B/14B SLA (#15888) 2025-12-30 14:30:22 +08:00
Stefan He
9416464682 Fix Qwen Next GDN w/ Radix Cache (#16053) 2025-12-29 21:06:06 -08:00
Cheng Wan
5fb734f1a5 Enhance comments in set_extend_input_len method (#16130) 2025-12-29 20:47:33 -08:00
Cheng Wan
60f1ca6925 Refactor: Moving extend_logprob_start_len calculation out of prepare_for_extend (#16105) 2025-12-30 12:38:33 +08:00
Vladimir Serov
9e263c2162 [LoRA] Torch native backend: rework implementation and updated tests (#15187) 2025-12-30 11:48:51 +08:00
Prozac614
f253f43c9d [diffusion] CI: fix LoRA downloading issues and respect offline flag (#15813) 2025-12-30 11:39:27 +08:00
Mick
1e45320198 [diffusion] improve: tiny improve layerwise offload manager by consolidating weights per layer (#16081) 2025-12-30 11:31:00 +08:00
Mick
26e17f9076 [diffusion] improve: tiny speedup qwen-image-edit-2511 by avoiding unnecessary calculation (#15896) 2025-12-30 10:10:32 +08:00
Ke Bao
3de23274ee Clean up swa handling in fa3 backend (#15877) 2025-12-30 08:51:40 +08:00
Kangyan-Zhou
9c4eb46099 Add a new branch cut GH workflow, and adopt setuptools-scm for version control (#15985) 2025-12-29 13:51:21 -08:00
Cheng Wan
c2e0913e17 Fix extend_input_len calculation in decode.py (#16103) 2025-12-29 13:13:24 -08:00
jiapingW
278012caa0 [Feature] support bench jsonl files with sharegpt format (#15057)
Co-authored-by: Xinyuan Tong <115166877+JustinTong0323@users.noreply.github.com>
2025-12-29 10:06:11 -08:00
Simo Lin
162d1cf9be [model-gateway] Add classification model support infrastructure (#16061)
Co-authored-by: Chang Su <chang.s.su@oracle.com>
2025-12-29 08:34:05 -08:00
Mick
8e08207c18 [diffusion] fix: fix serving with dit-layerwise-offload enabled (#16066)
Co-authored-by: Xiaoyu Zhang <35585791+BBuf@users.noreply.github.com>
2025-12-30 00:21:42 +08:00
Mick
b5d9fc873b [diffusion] chore: minor refactor by streamlining the VAE class hierarchy (#16069) 2025-12-29 23:37:59 +08:00
Xiaoyu Zhang
24616c5234 [Diffusion] Qwen image edit support qknorm optimization (#16062) 2025-12-29 21:37:24 +08:00
cklxx
d48723b77d Clamp logprob tokens with model vocab size (#14414) 2025-12-29 20:37:07 +08:00
Xiaoyu Zhang
f3d73b0199 [Diffusion] Refactor qwen_image's rope in a single helper func (#16047) 2025-12-29 17:24:26 +08:00
Xuchun Shang
4ab66d956f [HiCache] Fix deadlock when creating new group (#15805)
Signed-off-by: Xuchun Shang <xuchun.shang@gmail.com>
2025-12-29 17:13:04 +08:00
Xiaoyu Zhang
8305dc1718 [Diffusion] Disable packed QKV for FLUX & Z-Image (#16038) 2025-12-29 14:33:05 +08:00
Liangsheng Yin
a435f55d18 Tiny print launch command with shlex (#16010) 2025-12-29 11:26:46 +08:00
Mike Qiu
2ec6fa3c54 feat PD: add eagle3 support for DeepSeek V3 in EP mode (#14280)
Signed-off-by: Michael Qiu <qiudayu.qdy@antgroup.com>
Co-authored-by: Mike_Qiu <qiudayu.qdy@antgroup.com>
2025-12-29 11:12:34 +08:00
Li Jinliang
b840d6aaeb [diffusion] webui: reference to content task and better visualization capabilities (#16017) 2025-12-29 10:22:21 +08:00
EkiRui
ef4b3c0e96 Add host tensor allocator for memory_pool_host and support Mooncake standalone storage (#14873)
Signed-off-by: Xingrui Yi <yixingrui@linux.alibaba.com>
Co-authored-by: Teng Ma <sima.mt@alibaba-inc.com>
2025-12-28 16:28:50 -08:00
Cheng Wan
6f9d0a89a0 [scheduler] fix: correcting extend_logprob_start_len calculation (#15922) 2025-12-28 14:57:04 -08:00
Mick
d7a3336ebe [diffusion] fix: fix stages not logged when perf_dump_path is provided (#16016) 2025-12-28 23:17:43 +08:00
Xiaoyu Zhang
7d02c8e59f [JIT kernel] CI support jit kernel tests (#15939) 2025-12-28 23:11:02 +08:00
Lianmin Zheng
e6d5a213ad Fix metrics (#15998) 2025-12-28 05:03:49 -08:00
Mick
9f8e23071a [diffusion] chore: fix default offload setting for image generation model (#15928) 2025-12-28 20:45:33 +08:00
fzyzcjy
d90f9bfc4e Temporarily disable temp_prefill_info assertion to unblock CI (#16008) 2025-12-28 20:13:06 +08:00
Mick
3881bc8d0b [diffusion] CI: relax threshold by supporting different profiles (#16002) 2025-12-28 20:05:07 +08:00
Liangsheng Yin
be3828a13b Tiny cleanup duplicate code for multi-layer eagle worker. (#16004) 2025-12-28 18:08:20 +08:00
lif
5969be2f06 Apply fixture-kit mode to MMMUVLMMixin (#15615) 2025-12-28 17:22:05 +08:00
Liangsheng Yin
bf90ea9c5b Unify spec v2's naming manner. (#15990) 2025-12-28 14:14:52 +08:00
TZHelloWorld
0294844f04 [fix]deepgemm precompile when warmup (#15891) 2025-12-28 13:47:39 +08:00
Alison Shao
0e536600e8 Refactor: separate CI-specific weight validation into dedicated module (#15216)
Co-authored-by: Kangyan-Zhou <zky314343421@gmail.com>
2025-12-27 20:50:39 -08:00
Vladislav Nosivskoy
d70c265533 SGLang Tracing: fix attribute errors (header extraction & bootstrap span closing) (#15693)
Signed-off-by: Vladislav Nosivskoy <vladnosiv@gmail.com>
Co-authored-by: ishandhanani <82981111+ishandhanani@users.noreply.github.com>
2025-12-27 22:44:31 -06:00
Baizhou Zhang
656f4d69a1 Refactor fp8 nextn layer for DeepSeek nvfp4 checkpoint (#15353) 2025-12-28 11:57:09 +08:00
DarkSharpness
8e43980ebb [Feature] JIT Fused QK norm + qk norm clean up (#15835) 2025-12-28 11:53:50 +08:00
Mick
b4a00ed2d9 [diffusion] chore: clean ComposedPipelineBase (#15937) 2025-12-28 11:43:25 +08:00
fzyzcjy
f55d608c89 Tiny fix cannot launch nvfp4 checkpoint with bf16 kv cache (#15986) 2025-12-28 11:22:20 +08:00
Ho-Ren (Jack) Chuang
349ce2dd19 Support kv8 (FP8) with torch_native attention backend (#12596)
Signed-off-by: Ho-Ren (Jack) Chuang <horenchuang@bytedance.com>
2025-12-28 10:48:30 +08:00