 Dipika Sikka
|
d272415e57
|
[Quantization] Expand compressed-tensors MoE matching logic to support NFP4 + FP8 MoEs (#22674)
Signed-off-by: Dipika Sikka <dipikasikka1@gmail.com>
Signed-off-by: Dipika <dipikasikka1@gmail.com>
|
2025-08-27 05:00:21 +00:00 |
|
|
CSWYF3634076
|
644d57d531
|
[Model] Add Ernie4.5 VL Model Support (#22514)
Signed-off-by: wangyafeng <wangyafeng@baidu.com>
|
2025-08-26 21:02:55 -07:00 |
|
|
Michael Goin
|
de02b07db4
|
[Bugfix] Lazy import gpt_oss_triton_kernels_moe for mxfp4 (#23678)
Signed-off-by: mgoin <mgoin64@gmail.com>
|
2025-08-27 09:34:57 +08:00 |
|
|
czhu-cohere
|
2c2b140ae8
|
[quantization] use channel scales for w4a8 + misc fixes (#23570)
Signed-off-by: czhu-cohere <conway.zhu@cohere.com>
|
2025-08-26 18:23:23 -07:00 |
|
|
zixuanzhang226
|
b1625dbe9c
|
feat: add triton fused moe config for GLM-4.5-Air-FP8 on B200 (#23695)
Signed-off-by: Zixuan Zhang <zixuanzhang@bytedance.com>
|
2025-08-26 18:06:10 -07:00 |
|
|
Hyogeun Oh (오효근)andHarry Mellor
|
730d0ac8b9
|
[Docs] Fix warnings in mkdocs build (#23649)
Signed-off-by: Zerohertz <ohg3417@gmail.com>
Signed-off-by: Hyogeun Oh (오효근) <ohg3417@gmail.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
|
2025-08-26 18:19:23 +00:00 |
|
|
TianyuLi0andLi, Jiang <jiang1.li@intel.com>
|
f58675bfb3
|
[CPU] add cpu fused moe pytorch native implementation (#23146)
Signed-off-by: Tianyu Li <tianyu.li@arm.com>
Co-authored-by: Li, Jiang <jiang1.li@intel.com>
|
2025-08-26 14:09:17 +00:00 |
|
|
nvjullinandMichael Goin
|
f66673a39d
|
[Kernel] Added flashinfer fp8 per-tensor gemms (#22895)
Signed-off-by: Julien Lin <jullin@nvidia.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
|
2025-08-26 06:54:04 -07:00 |
|
|
Michael Goin
|
d52358c1e0
|
[Perf] Remove duplicated NVFP4 blockscales to save memory (#23379)
Signed-off-by: mgoin <mgoin64@gmail.com>
|
2025-08-26 19:16:33 +08:00 |
|
|
|
ae067888d6
|
Update Flashinfer to 0.2.14.post1 (#23537)
Signed-off-by: Siyuan Fu <siyuanf@nvidia.com>
Signed-off-by: siyuanf <siyuanf@nvidia.com>
Signed-off-by: Weiliang Liu <weiliangl@nvidia.com>
Signed-off-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Siyuan Fu <siyuanf@nvidia.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
|
2025-08-25 18:30:44 -07:00 |
|
|
Xin YangandWentao Ye
|
8a3cd90af5
|
[Kernel] Add fused grouped_topk kernel for MoE (#23274)
Signed-off-by: Xin Yang <xyangx@amazon.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
|
2025-08-25 11:47:52 -07:00 |
|
|
Ayush Satyam
|
5c4b6e66fe
|
[Attention] Unify mamba and attention backend selection (#23171)
Signed-off-by: Ayush Satyam <ayushsatyam146@gmail.com>
|
2025-08-25 09:09:36 +00:00 |
|
|
LIYIFAN_liyifan
|
c9abb10489
|
[Bugfix] Fix Dense module loading for sentence-transformers embedding models (simplified V2) (#23408)
Signed-off-by: FFFfff1FFFfff <yifanli0919@gmail.com>
|
2025-08-25 05:39:24 +00:00 |
|
|
Jee Jee Li
|
170e8ea9ea
|
[Misc] Unified linear print info (#23516)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
|
2025-08-24 20:13:51 -07:00 |
|
|
Ming Yang
|
504d914314
|
[Perf] Add Triton config for DeepSeek V3 FP8 EP32 H200 (#23504)
Signed-off-by: Ming Yang <minos.future@gmail.com>
|
2025-08-24 18:06:35 -07:00 |
|
|
czhu-cohere
|
e76e233540
|
[kernel] Support W4A8 on Hopper (#23198)
Signed-off-by: czhu-cohere <conway.zhu@cohere.com>
|
2025-08-24 06:18:04 +00:00 |
|
|
Chenxi YangandChenxi Yang
|
308fa287a8
|
Add glm4.5v tp2,4 fp8 config on H100_80GB (#23443)
Co-authored-by: Chenxi Yang <cxyang@meta.com>
|
2025-08-23 02:54:19 +00:00 |
|
|
Daifeng LiandMichael Goin
|
fa78de9dc3
|
Quantization: support FP4 quantized models on AMD CDNA2/CDNA3 GPUs (#22527)
Signed-off-by: feng <fengli1702@gmail.com>
Signed-off-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
|
2025-08-22 20:53:21 -06:00 |
|
|
elvischenvandLuka Govedič
|
24d0c9e6ed
|
[NVIDIA][torch.compile] Support Flashinfer TRTLLM FP8-q/kv NVFP4-out Attention Kernel (#22703)
Signed-off-by: elvischenv <219235043+elvischenv@users.noreply.github.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
|
2025-08-22 22:09:05 +00:00 |
|
|
rasmith
|
cc7ae5e7ca
|
[BugFix][AMD][Quantization] Fix torch.compile issue where wvSplitKQ not being called when it should when using quantized FP8 model (#22281)
Signed-off-by: Randall Smith <Randall.Smith@amd.com>
|
2025-08-22 21:47:57 +00:00 |
|
|
Isotr0py
|
4645024d3a
|
[Quantization] Allow GGUF quantization to skip unquantized layer (#23188)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
|
2025-08-22 13:04:22 -06:00 |
|
|
Wentao Ye
|
394591e343
|
[Feature] Enable DeepGEMM Linear on B200; 1.5% E2E throughput improvement (#23351)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
|
2025-08-21 21:01:08 -07:00 |
|
|
Michael Goin
|
3bbe11cc13
|
[Perf] Small optimizations for silu_mul_fp8_quant_deep_gemm (#23265)
Signed-off-by: mgoin <mgoin64@gmail.com>
|
2025-08-21 17:56:15 -04:00 |
|
|
Wentao YeandMichael Goin
|
48bfb0c9b7
|
[Bug] Fix R1 Accuracy 0 Bug (#23294)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Signed-off-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
|
2025-08-21 13:11:28 -04:00 |
|
|
Yi LiuandWentao Ye
|
0278f1ac3a
|
Fix nvfp4 swizzling (#23140)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
|
2025-08-21 16:54:50 +00:00 |
|
|
wang.yuqi
|
d70a16625d
|
[Performance] V1 Pooling Models E2E Performance Optimization (#23162)
Signed-off-by: wang.yuqi <noooop@126.com>
|
2025-08-21 13:26:09 +00:00 |
|
|
Paul Pak
|
2e2000f352
|
[Model] Add LFM2 architecture (#22845)
Signed-off-by: Paul Pak <paulpak58@gmail.com>
|
2025-08-21 09:35:07 +02:00 |
|
|
Asaf Joseph Gardinandasafg
|
3663870c72
|
[V1][Mamba1] - Full CUDA and Piecewise CUDA Graphs Support (#23035)
Signed-off-by: asafg <asafg@ai21.com>
Signed-off-by: asafg <39553475+Josephasafg@users.noreply.github.com>
Co-authored-by: asafg <asafg@ai21.com>
|
2025-08-20 20:08:51 -07:00 |
|
|
Li, Jiang
|
7be5d113d8
|
[CPU] Refactor CPU W8A8 scaled_mm (#23071)
Signed-off-by: jiang1.li <jiang1.li@intel.com>
|
2025-08-21 09:34:24 +08:00 |
|
|
22quinnandSimon Mo
|
4b795020ed
|
[EP] Add logging for experts map (#22685)
Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com>
Co-authored-by: Simon Mo <simon.mo@hey.com>
|
2025-08-20 23:46:06 +00:00 |
|
|
Michael Goin
|
0cdbf5e61c
|
[Kernel/Quant] Remove the original marlin format and qqq (#23204)
Signed-off-by: mgoin <mgoin64@gmail.com>
|
2025-08-20 15:13:36 -04:00 |
|
|
shixianc
|
b17109beea
|
[Kernel] CUTLASS MoE FP8: Integrate cuda moe permute/unpermute (#23045)
Signed-off-by: Shixian Cui <shixian@amazon.com>
|
2025-08-20 10:35:26 -04:00 |
|
|
amirkl94
|
a38b8af4c3
|
[NVIDIA] Add SM100 Flashinfer Cutlass MoE fp8 backend (#22357)
Signed-off-by: Amir Klein <203507526+amirkl94@users.noreply.github.com>
|
2025-08-19 18:01:53 -04:00 |
|
|
bnellnmandMichael Goin
|
b94faf9d50
|
[Bugfix] Fix accuracy issue when using flashinfer cutlass moe, TP=1 and modelopt. (#23125)
Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
|
2025-08-19 14:00:51 -04:00 |
|
|
|
1298c67795
|
[FEAT] [Performance] Enable DP for ViT in Qwen2.5VL (#22742)
Signed-off-by: tjtanaa <tunjian.tan@embeddedllm.com>
Co-authored-by: DarkLight1337 <tlleungac@connect.ust.hk>
Co-authored-by: Cyrus Leung <cyrus.tl.leung@gmail.com>
|
2025-08-19 15:25:57 +00:00 |
|
|
wang.yuqi
|
f856c33ce9
|
[Model] Add multi_label_classification support (#23173)
Signed-off-by: wang.yuqi <noooop@126.com>
|
2025-08-19 12:54:30 +00:00 |
|
|
Chengji YaoandChengji Yao
|
e9d6a3db69
|
[TPU] make ptxla not imported when using tpu_commons (#23081)
Signed-off-by: Chengji Yao <chengjiyao@gmail.com>
Signed-off-by: Chengji Yao <chengjiyao@google.com>
Co-authored-by: Chengji Yao <chengjiyao@gmail.com>
|
2025-08-19 11:46:42 +08:00 |
|
|
Michael Goin
|
6d25e3fd6e
|
Use Blackwell FlashInfer MXFP4 MoE by default if available (#23008)
Signed-off-by: mgoin <mgoin64@gmail.com>
|
2025-08-18 15:25:49 -07:00 |
|
|
Simon Mo
|
0fc8fa751a
|
fix: gptq marlin weight loading failure (#23066)
|
2025-08-17 15:56:07 -07:00 |
|
|
Jee Jee Li
|
4d4061b6e7
|
[Kernel] Add cuda kernel for gpt_oss activation (#22951)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
|
2025-08-17 05:03:24 +00:00 |
|
|
Michael Goin
|
94096a47c9
|
[UX] Separate marlin moe config logic from triton moe (#23006)
|
2025-08-16 22:16:42 -04:00 |
|
|
Jinzhen Lin
|
a258ad8bcc
|
[Bugfix] fix qwen3 moe fp8 accuracy issue (#23031)
Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com>
|
2025-08-16 17:41:23 -07:00 |
|
|
Michael GoinandWentao Ye
|
4fc722eca4
|
[Kernel/Quant] Remove AQLM (#22943)
Signed-off-by: mgoin <mgoin64@gmail.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
|
2025-08-16 19:38:21 +00:00 |
|
|
bnellnmandMichael Goin
|
8ad7285ea2
|
[Kernels] Clean up FusedMoeMethodBase and modular kernel setup. Remove extra arguments from modular kernel methods. (#22035)
Signed-off-by: Bill Nell <bnell@redhat.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
|
2025-08-15 14:46:00 -04:00 |
|
|
sstamenk
|
6b04039a72
|
[BugFix] Skip the Q component for QKVParallelLinear in the case of QKVCrossParallelLinear since its width is 0 (#22369)
Signed-off-by: sstamenk <sstamenk@amd.com>
|
2025-08-15 17:17:31 +00:00 |
|
|
|
75531a6c13
|
[V1] [Hybrid] Support using float32 for state in Hybrid Models (Mamba2, Mamba1, Minimax) (#22928)
Signed-off-by: Daniel Afrimi <danielafrimi8@gmail.com>
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
Signed-off-by: Chen Zhang <zhangch99@outlook.com>
Co-authored-by: Daniel Afrimi <danielafrimi8@gmail.com>
Co-authored-by: Burkhard Ringlein <ngl@zurich.ibm.com>
Co-authored-by: Chen Zhang <zhangch99@outlook.com>
|
2025-08-15 12:57:06 +00:00 |
|
|
|
fe91ce9591
|
[V1] - Split Prefill and Decode for Mamba1 models (#22653)
Signed-off-by: amirk <amirk@ai21.com>
Signed-off-by: asafg <asafg@ai21.com>
Co-authored-by: asafg <asafg@ai21.com>
Co-authored-by: Asaf Joseph Gardin <39553475+Josephasafg@users.noreply.github.com>
|
2025-08-15 08:59:52 +00:00 |
|
|
TJianandvllmellm
|
b2f6c247a9
|
Revert "[ROCm][AITER] Support AITER Rope ops in RotaryEmbedding Module." (#22956)
Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com>
Co-authored-by: vllmellm <vllm.ellm@embeddedllm.com>
|
2025-08-15 06:39:19 +00:00 |
|
|
Asaf Joseph Gardinandasafg
|
3d232dbd19
|
[Mamba] - refactor: Renamed mamba_attn to mamba2_attn (#22818)
Signed-off-by: asafg <asafg@ai21.com>
Co-authored-by: asafg <asafg@ai21.com>
|
2025-08-15 06:38:05 +00:00 |
|
|
amirkl94andMichael Goin
|
b4cef5e6c7
|
refactor: Change scaling factors calculation for flashinfer FusedMoE (#22812)
Signed-off-by: Amir Klein <203507526+amirkl94@users.noreply.github.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
|
2025-08-15 06:19:31 +00:00 |
|