kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-02-03 17:52:19 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

[https://nvbugs/5783509][fix] Fix a hang issue when enabling skip softmax on Blackwell (#10490)

Browse Source 
Signed-off-by: Tian Zheng <29906817+Tom-Zheng@users.noreply.github.com>
This commit is contained in:
Tian Zheng 2026-01-16 18:59:54 +08:00 committed by GitHub
parent cc43edc8f4
commit cfebfbb505
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2007 changed files with 5892 additions and 5970 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvCausalP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:aab9ac4e7321815981e7f16588880c4471bc11d7ef847c5ba59823a61f7102c1
oid sha256:17f34a6d5d51c8e1f9ff6e7214de27fde634db43225c384c2587fc9462f1748c
size 630932

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvCausalP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7c42a76a6d469e9981fd31517a0eea89ad25176eafd9ec847a90965e69e8da58
oid sha256:bf58b9c01ac2b2ead1fff8b6f77f831df63c6c4e7f7e53dd6bca2d491aba5f62
size 565951

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32MultiCtasKvCgaVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6d203f17a13605424fc336835b0da020a03b7822a2eb57f3b49180b26fc867c5
size 485079
oid sha256:4ff262f0ca67a42acdad47466eef7927fc160e86bf419eb1c2dad23eaa124158
size 483501

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32MultiCtasKvCgaVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:26f4f24197d75d46fcbedfc9de1557a994800d47ce7b0c22cc04cb3927bc7706
size 454565
oid sha256:5606453cb0b2ad5cc87fe714b9f3db12149a074cd942e3e3fcf81e39b1b8e435
size 452197

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32MultiCtasKvVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d79548d5a0ab81b8bb4ab36aefda378bd908fff8be0bcd6b3c6363e29b5743dd
size 472765
oid sha256:01b753b3b47322d84e41f3da336fc412e506a07896f404b698995e1f26a3ceb7
size 471975

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32MultiCtasKvVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:344abd862e54d481b3c9edcd16218fef31f7a443fd7f2a9a839de743f6738145
size 450169
oid sha256:09964a2485eb822cd27cb559014c80200cf3dd779a10f94a547f76e6ca795408
size 447801

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2885aa27b1f8efc168f3ebde84fc86678ab384e2e35ef5a3b208289e94690ed7
oid sha256:f2b406054d3d62e07d36ce3c9c81ea7d91a7b5a8a605ceecbe4ff0c7fa4ea825
size 626982

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0270860fd490ab53e5a9fd28cbd4219ccaa4375ddd9919fe721dc1c6b3243a36
oid sha256:44398e40e44aa59b773ddc47980d87e4e7db7f2f8bb79fa1cd47c94ad1c5a7a1
size 566023

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32VarSeqQ16Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bfd79f0c8f79be828198380b0b1a0868d3168a447659dca8752cd9dd732f4503
oid sha256:359a7f91fabf1fabe222aa4204b35726a2a0bbf4f78d3fe1233e9dd1dba6d37a
size 452573

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32VarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bb30dc1e4c550621636e8c853e6b104723667fd2b48d5674c85f1c4617fa874f
oid sha256:6ab77904c1ac7c61e8244fdd8693f0284152f7ef256a467d41576cd0ca4f9b70
size 394599

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32VarSeqQ8Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e97c1dbc7556316ec7e35f078732736009c83e29fd7dffc0b9bb1449eb1da0f0
oid sha256:a977331738ef6f94c056fb0415067677577a3ccc0b4f5b35ba9f1af81ecb1da2
size 436291

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvDenseP32VarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6ed203011340282225246e72220500aa1d555f68036b786fc2db0ee0fe98806d
oid sha256:c8753486471497fcd14296cbe79ed0217fd96c736e8bcaae2ba7f64e86b56f17
size 378317

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32MultiCtasKvCgaVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0c719151d9ec920ede687ea6e643bf6f8b4fc51d298c23dac9a9a480a9abfb4e
size 508005
oid sha256:5e86ee3fcc48ddae27a34f84cf83a7c61a634a27277fb96b33e75cd5c78f6a52
size 505637

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32MultiCtasKvCgaVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cf13e09558450edad962a3b2dcb40b241e19ae73fc094cb5e3b4d2e1aeb0f4ec
size 471175
oid sha256:356c67ce617da2547bc3969fa893638bb1b824684db7eeada2b64713c69e0dfe
size 468807

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32MultiCtasKvVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c94d9acd5903fc23f77a8f9a50725cd3702646d8e4cd1f8ac347bba2b71bd579
size 495689
oid sha256:ddaf641d7ed2bf7bc20f5a84bfb4b801297d83d00baa320f1b38948bc575aeb7
size 494111

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32MultiCtasKvVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d7afb5710156b1107f18c149ffc429770348d387c0976fdae13998d895ec5eac
size 471515
oid sha256:d35dc6e393dbb3e06a8ded38d3beff7977d4b768192ece6fec69939df0d0454d
size 469935

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8141ce31fa80e0b34ee5103cc88dd081b9e46f9190a2c1417e0a76b852069bcc
oid sha256:3404acdcd673f67dbd15f247cfdea00c87f19370defaba5884b5b9174fa596fa
size 670528

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d6d3c73fa9bab18a2fd5640d585e6456cdbf2cd2ce0c768c6dc96db7f4341565
oid sha256:5d7fd84cc238fb1002ac9fbe5ffecc1dda3fa40ac028f68bbca44b5a74017aeb
size 607571

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32VarSeqQ16Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6fb4e40e49c327b4412be9d80d2fa36f69ee9b0f9f3a189cb69d231ab429cf52
oid sha256:bad8752b471b77f4a8343f9fdb4127956f55ff1aee21b2c0d606daebb1083880
size 481715

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32VarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f5c464e1d6055c7b1b914ba28ffd46214dd067074f1ec5845c301dab56fef3b1
oid sha256:70255b5c11bfb4e06de22f9106aecb070d0a336568b121ac3f9bdc59ac8ad9fa
size 411875

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32VarSeqQ8Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1283f9838b33699479b3cb220ae7a6daeaba289df8f6784da22226b65365c04c
oid sha256:a20420229225ae1c2c0e853a0a585d04793e435d63cd6090eb7e38d9cbbb18e2
size 460697

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H128PagedKvSlidingOrChunkedCausalP32VarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:197830e1331734f431c27ccefbc6b8abd43b5cbdca9633ceff01de653409f37e
oid sha256:8b43826a9a48541cba45ff0082283e8a069cbbc9894dae055a2402cbccef7777
size 394015

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvCausalP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:44787d6712a551d25b917d99a6eb5fa13500ed03f2e26a279b341c88c4140902
oid sha256:0fd569e36929c1a54ec573f5e1f53d65ce3630c5689920fad2864422f7dea7e6
size 663022

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvCausalP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:07508bc40af84a09327533065d5a092465a5c2f206165c57a36f5149bb898804
oid sha256:af6d5c8888d51e54ffcdb05f4c42016673c684c878f901be8025edc6e490b2aa
size 574435

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32MultiCtasKvCgaVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f7a631d93550a5c4649669883cdc3bce68b0573ab5a6dad93f6748647ff42180
size 595587
oid sha256:0bd694d0ba8cc061c6932b070779d57815336f59075d189f57ff010103a621c7
size 593219

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32MultiCtasKvCgaVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7165266f1f1f613f44ce22e05e4953348c3a75f32b7652ce14b21d6b76f0d3c0
size 571387
oid sha256:d41cb296ae0343e7f017be89295d49233757f5cac6edf07c3cc96cd6bb907bf4
size 569019

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32MultiCtasKvVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:318e2c4385be1e2569c5702c56b66d9b8e223bb4bebd1e15eef9dbcc8c0f9029
size 581693
oid sha256:a88239db53ab73c5589f06469a4afba6d20c1f23c8033e740a628d970697d93c
size 580115

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32MultiCtasKvVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3131c8d5f5fb015fcacd145c3017473825baefd64a60a624f3289030e05104f9
size 560675
oid sha256:5049e27f0756a25106f5615d60e539bc514a798a722146380622b8d2ca32694e
size 559885

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:024dc41efed33d2c38c11d18255009a9a3b2cc2800aee0d8aa81ad635448813f
oid sha256:1ab33d785fb970903081deca276f6904aa4a58101b17839eefb96f6f324af263
size 663760

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f078613e55c6832616133eef2f68be51cca4e4a4a048edd2f8630c98763e9d64
oid sha256:aad7ae4f3f4dd336159cbd10009f072359d428463d552cd478d30fdbc7e936b5
size 608499

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32VarSeqQ16Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e018fa3ab404f4a2af235291ff638dcfbaee86f48a6eb2fff84b8642fd64c804
oid sha256:57c8aff9083d68a7979c626e6ba095eb8e48b61fddceddf3721a65f71877be34
size 558345

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32VarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cbdbe20d955c73deff21abfa6107d8d49cc8560f5eb22d9ea9218ccd08ca3d5b
oid sha256:a566742e538f476775da8d6bb0566f21bd9b8c0491b427d4818a450faf2d8148
size 501159

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32VarSeqQ8Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a14260c98eb9cbf4cd991370fb702da342a564a2090d0f2ba85917f20e801f11
oid sha256:50402fdd3ed68147dfe0b021c6c8b2515a00332795734f9451ce39edd666573e
size 539695

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvDenseP32VarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4a92a1759f8ae7d760193a645c93ce7bfd54681a800839a0d8ddbc9ff8ec0ecd
oid sha256:500a9236fc5e897b2c406d755def2dbc67cabb00f097e4b8c34419511cab2f99
size 483299

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32MultiCtasKvCgaVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:39712104e975e87c0a6b4a7702a9adebce0c13d429271ed20c0e461c57f0535b
size 618512
oid sha256:b2a8fade86e5093027c001b27a612a5e7ee18ae143e08ccd00043f4bc32ccdca
size 616933

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32MultiCtasKvCgaVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c133abb655b3c4951333b48115f8641eeb4d380245c20565fcd5c0ef55c5c4ff
size 593521
oid sha256:28bd8634521092266655956342505e5f978559a41100e22c5fd0c801855bf166
size 591153

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32MultiCtasKvVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:fad53d9ba155ce085e9933b5b5dfba6fc07e153d65cb200a26c3a4a5bc6199bf
size 604617
oid sha256:e5b51830511bdb82a252cd25a06167e7f97bc160e80e8104ce556ae37229ff67
size 603039

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32MultiCtasKvVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:aa632239be24a5a2707fc6f6e85bc83479072e1294cd9eac2c8fb47993c6bd16
size 582811
oid sha256:9840bbba2d04a1c72e61e7c9bff213eea3a1068ec19998e34cd5916e0b9e0f3a
size 582021

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ea57bea5a2fc471860e4c5525fdcae29470dda4fe1300f99036585586d35a140
oid sha256:19f636770a8f6b8c2112c8d6c1eb8d991207d7ee9f67b4c8c18a7f0028602b6f
size 702670

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c91939f73ae90335e7233a1b8bd732ed3cc94d178172388bab60bb7d01aeed9b
oid sha256:df92ad291cf78872f979709e07e4a85b1647a34bc76dd53ca31d71ce08e62b37
size 617191

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32VarSeqQ16Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b46ecbb25482181e34dbe16854abe1c5a688608834821c5679c057c5fb7c43d3
oid sha256:c7b6b8e819ec825f331ed01b7d9f7d7fe115050ca29fe6c3d31e87146250d56a
size 586695

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32VarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1bd636e50b2cae9c7cfb5da70a1f59545cf5efef7ecd00f9036f6dce48e2b98b
oid sha256:8bca2490575bb5d45ac37a176b7da45fd91445e0ed058bb8aea0234941289cb2
size 512909

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32VarSeqQ8Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a1894d55f3eca9f0d9299fb9704a07ed69e3dfc9c35cc6e2e5b2da2dc35c6e97
oid sha256:527acfcd515e35e2a383cddf972f2b7602a381327633ba20d6732319cb6bce98
size 564889

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H256PagedKvSlidingOrChunkedCausalP32VarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c98aec7849f9db765d3edfb979f219aa5419466b9e1e54eea67ec9f5df58c0bc
oid sha256:913334b768bf0af735fec8189dd0806f6b22f40a0cd30e10a538ce566237e7e1
size 494259

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvCausalP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:83c25ec85687ea926e665fbeb370dec55d36412238318013b0a392f87fb577fc
oid sha256:1e22d2b124c9d85ed319e17a5f36c6ffb46e47f4c520fe9c31c9fb46fd6a118b
size 561491

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvCausalP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:75804fb6424217fcaf15ce78a87c0629ecc35b6a3a1f43eddad1354691c581af
oid sha256:2ca4fe139a26e0bde389ae0dfa2a3a76ffe79367562bf7ba454e5936808d6107
size 502949

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32MultiCtasKvCgaVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6e86ec832ea89bd957499fcbf2d38d86b8dc3170e0e657cd94d83fec34f0a155
size 459251
oid sha256:1dfb56733642035a86b7ddf56294109ee185f122e2a69a493797839734ddeb4f
size 456883

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32MultiCtasKvCgaVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:be19e95ebd0d58474840670c8d616b56eeebde025e3b4853bd59f206ec5666a9
size 444301
oid sha256:a776f2051102716638f00e252f1c20829719c386e9cdef9a418368996454320f
size 442723

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32MultiCtasKvVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:524a093590794c4cf8f42cf6e6fa639bfd5e6d86be5050f0f538341f9937ef5d
size 453227
oid sha256:356f207596ef727c1b12c4a01950869ebd066b583b6b2a7b9423c9804ee317dc
size 450859

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32MultiCtasKvVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b90becb8d6b7abc262309e319f7fa6a6acc46946be96ced0219f1a762544776d
size 438327
oid sha256:9f2035c671f6a86d4622a2061df21684f05c74d23e0c6c2d92f869b362dfe6db
size 436747

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bc0433873f87e978d8ee8262d23a8726e469935fbeb3266861343b42f4cc9f4f
oid sha256:08a3c22687c7327734fd6c0013e2b2cd19003c29f72b89a5b5c53835950ae91b
size 558333

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4619457c5cca7dfec5595fe210944782362dcb2967cef79367dde603f11dbe8a
oid sha256:5f93ba7d2bcff0f78486277efe6f1c13fe2387aa70605b916cdd22fb15ce1c31
size 502233

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32VarSeqQ16Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:559977ec6b2d1c414c49bdbed7059f355a7c28b71bdc155c8d558a6895185884
oid sha256:61d7b6e0b56fbe8d6b9cca3cb934f710391654765e76f6f74485baf525f50c8f
size 441719

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32VarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bc4fe524bd66420e82cc803d3afad4c9953da3d03a3bc8fc76afbf479c5fd820
oid sha256:434de3e33d46200c6aab0f120c05b16b2b29bed1f6f7d06a309c0b3cb2c9fdc9
size 384533

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32VarSeqQ8Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ba6a0d2ca4cc7230968f4e0566854f8a3e3b1ea7034bc835f1c72a344f5840c4
oid sha256:09d25663c25667d1ebb2a504ba6c261c87d77fa512d15232bc80cfbc65f5eb7f
size 424449

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvDenseP32VarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:776e131f5d7eade5570dfc0209e09fafd2130585f4e401278d9985a1c5a75923
oid sha256:17fa99b7bc15154573aaee6e6c0aaa918c6c959daec291cdade13e5152479921
size 367265

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32MultiCtasKvCgaVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4dd17b59709ffef17fc30dbcf6116ec96149628f7bafe78ccbeb77140b8969d9
size 475813
oid sha256:5d42e7ee8b11a2c06cdcf012826ee0020ba83494713d22225d358d2d90f9eccd
size 473445

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32MultiCtasKvCgaVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:9e5726b4f5372a563113f43a160f589d1a220ac545510198b28b93911d715e4c
size 460911
oid sha256:c4e5bd42ce6da3b5d4fba10b39ea5f6d755a103b9e1b6dfb1c43e942aded5850
size 458543

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32MultiCtasKvVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b90eb1e83de198440c6f74752615c936ab362cff2292930b90e1298b2267fb34
size 475363
oid sha256:63ee26044f6e3cdf5558f438c2afcd6a17a3a0208ff4635111831a8108d85075
size 473783

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32MultiCtasKvVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:73e9a3cef949138e3c137a8e1aa629d4ee4bf66951396556ba87f220997616f0
size 461251
oid sha256:6ddf4863648d25cb764f8dc0a0c1f7fb750d128fb6f49100bf96668d2ca75079
size 458883

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c2e9cc72bd2046f55d1f60d5b6211f205214d41feead79ce7cb5f7f73aa65e1d
oid sha256:4955e59ea224d35862cdea6a1340bcdc1d7e26cdd0e9063b69dfc0bfac58e43b
size 601089

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:316a360022761cb9ca2e6ddb743eae3e358c4264b23dfcb892b7dfa266f2981d
oid sha256:85567463b222f06d5dc0de196753ac3e2675851a63010b7b0db0ac6a7b003b9a
size 545605

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32VarSeqQ16Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f57627902c45ef259d2bcce82d2aed91f4f1e3138b4c96f72a123308c2690654
oid sha256:26abc8741b5261ec9a388a4f13d50a1fb42fbc1891b50cd226c66ad9a0cdfe3b
size 470069

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32VarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7d1c703c29fd454facd4eb8157d517023cbff84327c029da1ee450fc384177ba
oid sha256:ffeb03d13ffd0586579097348aa7622db56b1af2ab427e39feded5a85412bae2
size 400231

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32VarSeqQ8Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a2faeeccd23b935262318aefa1537166136d787ecd0162724e74551036766db6
oid sha256:8bc137f117ddf4772b21786a2e244ccca365380a96b941c99168828126cd51ea
size 449643

2
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100aKernel_QE4m3KvE2m1OE4m3H64PagedKvSlidingOrChunkedCausalP32VarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5ca0627fb8f03a86483556cea780be7fa229d0e6f6a41b65a6bbae991e26e7af
oid sha256:abb74fdb16d0939e6f8e2ef6c9e210e2605c4606a11b356ab111bc8cea92b3b2
size 382961

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvCausalVarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:231c5272ab1d7866af9cad01a1fe4a40977f115e2e913df23a71cdbdb28a8bce
size 727038
oid sha256:ab4b2cbfb75c57ab5e13b6f7b83ba8bb242a7e467e72b39aaf434c4bc7c6aead
size 749338

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvCausalVarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8643edd12e91ff89c7c4454df5cf63b75e85bdd77cab630cb6ef40353bb6a968
size 639218
oid sha256:07064296444d7ac6bfff94dcc790acd605fef68a760cbc5ead3296c880789567
size 661516

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvCausalVarSeqSkipsSoftmaxQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:74f68c681bbd2d78fef4264e317f336742785846ade3413e657c35d10ec77c6f
size 754542
oid sha256:f4f5ea892f241b754a11f09517c7ba065046e9b65c35ebcb0bedd84c745eba9c
size 776790

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvCausalVarSeqSkipsSoftmaxQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a4b6e4c1ed3c63c056494ab38be3fcae331b53e80cf710c24e98a38f05f5b3a4
size 666178
oid sha256:b354769aa9f5659b436534f7a3cc4aa09a493858d43ab70bdc316f6b0d88d512
size 688476

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvDenseVarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bfd95f9211563b3b7adaad3e0c404e7e59b38f9156b92769483017a0f605d082
size 730984
oid sha256:848644b2465232f39be0f82ee864ceb30e872c5a951bfcb80ee52bdd7d51bc13
size 753232

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvDenseVarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:209aea8af35e234dbbadd77383b20ae44cade2560ad72b5ac1e1c8d03e0f0bd7
size 639560
oid sha256:b35e56a59c42ea0b6f30bb49076c0f93611fcb00cdd76dd4353b465adc0c75c9
size 661810

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvDenseVarSeqSkipsSoftmaxQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:dc0e68ec4ab38cc24ba5389f4c1b7545cf40970dc8cfd8257ba7b583d7c96eb6
size 751036
oid sha256:a079ac7c297f2bcb5b6e2008140e2669ee3b50d2d49f3ee8b78cef9f9e80c33d
size 773336

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvDenseVarSeqSkipsSoftmaxQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ef655a211302525c2f15c128c7f3533879d2bd58d67061543332bad2ccb0ee95
size 666274
oid sha256:0f445f0ff16b08249c7700137d350ee6b1964ddb183bf4d5646852504a38dc61
size 688524

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvSlidingOrChunkedCausalVarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:08741252964a383b578df524498d5f01184f7b24974217d915b224b19cba85bb
size 795792
oid sha256:ed32d5e1a399b19c5b5a951b51efdd725591ab5910e434e47b9219ed9e8de053
size 818090

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvSlidingOrChunkedCausalVarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:aef6f373a556bea3541cf7d0ff9ece8b53543d21b393510ea9304635ff97b8fd
size 709450
oid sha256:80eaf7e695a9144f3e9ae58526dba4d62e19c374e3e3f5191d4216e2276ce68b
size 731700

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvSlidingOrChunkedCausalVarSeqSkipsSoftmaxQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f2b730c4d6e5642427c570370f0b0dc05117c98f3723e7d5d441f057361d982c
size 822702
oid sha256:269abd26b450b88d2fa3ebef00b8b50c9bf7acad9f49d8d7aa9dede9cfcc5297
size 844952

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PackedQkvSlidingOrChunkedCausalVarSeqSkipsSoftmaxQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ea07186733a5dda908eac7c77961577c30ddea22ce9b19166ca7e0489bc3a3b8
size 737742
oid sha256:9de7275c32c73482f2ae5e5336051bdbc172dfe02cd35950535363775a97636a
size 760042

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvCgaVarSeqQ128Kv128StaticKeepsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5666bc9080d54a4697174e95af03f17178d1d325093b154fc6e3307e9b1326b2
size 857510
oid sha256:5be32375b4f0595bfd691d4b01e5851737bda01669dcbd1a101db1e7fecb7f8c
size 874036

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvCgaVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:44c6b798b76cec7121f1d355ed73c98a1e610a51ba05dfe93ed5d0ce6cb42f79
size 706202
oid sha256:83558fa89e1d26bbf51093c42c3cc3af31711cb555fb496620a95deed45345a5
size 706006

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvCgaVarSeqQ32Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:71bdb382e0919e6ca38fd49fe3f3f939f9fd6392fb94ebea980c39b0557ba72d
size 774282
oid sha256:6e9ca086d9b2d1a1996b6b948efaeb61c609f8c052b901a1dee4e3992220784f
size 775714

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvCgaVarSeqQ64Kv128StaticKeepsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:df3c8bce1778c6dbdd406d3d8083eacd6e5a8674680105e793c1cdeb27fefba4
size 766192
oid sha256:9849935550b57458ebbc1d8cc12f4db1b05c33af085aedd5df63b3c845ad342c
size 773198

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvCgaVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:57de5817297d6e8a0624ab456c30f30626a5677fbe9df79640883d4c1bcfb008
size 661356
oid sha256:42e8365c84d53022a6ef4ecfc0ac76a079fb3b4637ee7ebdb43d2f16da0f12e0
size 660370

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvVarSeqQ128Kv128StaticKeepsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e0a28e454730e2518d74a00993124a1d575ac591d5d6691813a7f0bc93b0ddc4
size 816952
oid sha256:f2987e51b51870e7bcfdb3e631f4cbd0f5fbe47f9a8fb8f219fb6f716d11c27a
size 834318

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvVarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:03044da2c5640e8c810a9c8ab458749d971517d1dacb5093555598261b205573
size 692728
oid sha256:a05993e27df6d84f3782bc18b4d82ca8b4560c3c6447fa0d18a549f7206b42c5
size 694160

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvVarSeqQ32Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:fdaa0bdb7b8141dc2b4a26c8b0e610a419cabd41dd8c3deb66de6629ee771ba5
size 750152
oid sha256:7d87ad519f98c10ec7bef4c9c77e6d7dd8239778005f7808302903d900ce00e7
size 753162

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvVarSeqQ64Kv128StaticKeepsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:9ad167d9e27d219b65e5ac4d9e9ea0fe67f56659266a0da1483171fb63d639be
size 706394
oid sha256:8180e747e241a2a539db4613fa33edf0beea927ed8c8fb72a4afcdb1628ac841
size 714978

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32MultiCtasKvVarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8518fce896494d604e4f0338939b1e51e5210b946765adf42ea43ef0f409b930
size 655430
oid sha256:91f1f04bb85721ed4b00dc30e63ad0facbaa2d72514237fa2726adc9f5f78701
size 653654

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ128Kv128PersistentContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:9048c9448983013d1bb444eb0f5d04d5da27293625e825918ad135ac37920e2e
size 826448
oid sha256:3f9bc3320af971ab6c75b95a8f0757f39996222f5ef84735cc7ed5ca2f87e7e9
size 848746

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ128Kv128PersistentKeepsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:43395e3061fd8d67f3440c42bd6e381b3ca9d84196060e1a0657b63f327f40f7
size 806874
oid sha256:d2efa0d321c999e39ba1aa4ea2ce76b9298cda87df4c5e9ff5e14cba99cae803
size 825966

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ128Kv128StaticContext_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:037eecd82f5b89666d3644852a6918932af20127b74131f9caf33951c13eaf07
size 732608
oid sha256:fbe104d0f5e5db983d836054f35cd13ae589d82bc1037f0affb874bd7dab3648
size 754906

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ128Kv128StaticKeepsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e6b659558b18eb5f0eef9877821a185ee92aee7b984d03c28f5a32eac7335d26
size 711702
oid sha256:b90e116f2122aca36640ab90f96702da4462bc2736ba0665e57fe5d39864c34f
size 730794

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ16Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b697a2b0514c0d560826fb70c4fe0311f9f386a67cf9bd53323668586d9ac259
size 717430
oid sha256:e9aba7ec5f3cc7c07a1727ba7a453b8f8f502e1ace7cb8342f95ba155c966642
size 719058

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ16Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f2db5c67a62844819838c54fea4e6e342e7c4f9c93d0f5fdbd25e015852fa1a0
size 608889
oid sha256:dc5ec9515f97246cfa819c8612db5b0f61a11ef92664d88b376677dc83b16f9b
size 610517

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ32Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5040ef8f4c30f73184342b04d0b6d49768722fba6e362c43e5f30ca338bd3152
size 802530
oid sha256:57f9664d0518bd05aa601b8500f3f1043bcfe15d6c6da83393ca8147046dcec5
size 805738

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ32Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b2382fa3168de022fa01a7583e59f158d9ad5ce72b0429fbce3fd3f2cfa8c701
size 688020
oid sha256:590f5ee4d4d193f83cdc6a53206a928c1f0eeef0d5f8ccac351996d91ed6cfb2
size 691178

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ64Kv128PersistentKeepsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:271b915059fefb5f4a40d88199cdd9c6fa9a5572e94db17a4d01bac2dcb811ce
size 711412
oid sha256:23c707523597fdece25c21a724397f58fe46b433cca3b2c2e861f19f53370332
size 720982

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ64Kv128StaticKeepsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3537f75f09ea125568e8f120727f4bbe0fad01677e73606e200d8206b299d4e9
size 609777
oid sha256:0e01beb1a45556a3c584ff9fcc2d5a8b73a131104e7a3ca4221e7bb486dc9d95
size 619298

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ8Kv128PersistentSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1d00bd25788cdb73d285cab7f15915eb26653a433b643c7a57a6202763f70279
size 657440
oid sha256:eca4a5d71eafe647fc4a9f233b07279bedee9e91fbbf59c2492728a23a4d3a77
size 658228

4
cpp/tensorrt_llm/kernels/trtllmGenKernels/fmha/cubin/FmhaSm100fKernel_QkvBfloat16OBfloat16H128PagedKvCausalP32VarSeqQ8Kv128StaticSwapsAbForGen_cubin.cpp
View File

@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f785ac5a8cb1e160c0637235a68a3b6bd67d770a016ff40b76774c0b7a5069ff
size 564733
oid sha256:00f733e8db3ce48fb833a45491c678f5c754398fe3122b64d46b84a52d431017
size 565571

Some files were not shown because too many files have changed in this diff Show More