Skip to content

MERGE COMMIT ONLY, DO NOT SQUASH: record v0.6.18 as an ancestor #353

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
demandal25 merged 1,126 commits into from
Sep 2, 2026
Merged

MERGE COMMIT ONLY, DO NOT SQUASH: record v0.6.18 as an ancestor #353

Changes from all commits
Commits
Show all changes
1126 commits
Select commit Hold shift + click to select a range
b9890c8
fix: move JIT symlinks to writable gen dir (#3468)
namgyu-youn Jul 24, 2026
b190523
feat(comm): Add FP8 quantized two-shot AllReduce via symmetric memory…
kailashbuki Jul 24, 2026
fbbd1af
fix: let FLASHINFER_CUBIN_DIR env var override flashinfer-cubin packa…
ianliuy Jul 24, 2026
abe61b6
feat(moe): add unified per-tensor FP8 support (#4091)
feih-nv Jul 24, 2026
817e4bd
Update MoE CODEOWNERS (#4127)
feih-nv Jul 24, 2026
cea7f46
comm: multicast-free (IPC) workspace for TRT-LLM AR fusion under Conf…
elvischenv Jul 24, 2026
80c8e70
feat: fused FP8 quantization output for LayerNorm (#3962)
elwhyjay Jul 24, 2026
3e7106f
chore: Add Anerudhan to comm codeowner (#4135)
bkryu Jul 24, 2026
ec2e0be
CuTe-DSL modular prefill: sliding-window support + band-mask performa…
pgera Jul 24, 2026
3caf5d2
fix(gdn): compile SM12x CuteDSL kernels as sm_121a on DGX Spark (#3960)
nv-tusharma Jul 24, 2026
fa2672f
fix(moe_ep): fix two NIXL-EP combine deadlocks under real serving loa…
Anerudhan Jul 25, 2026
290c091
Adds SM107 support (#4122)
Vinnie6167 Jul 25, 2026
4b969c9
feat: close feature gap by wiring up per-tensor routed FP8 fused-moe …
jdebache Jul 27, 2026
6258e52
Revert PR 4122 (#4171)
jimmyzho Jul 27, 2026
f057e15
[GDN] improve sm100 GDN performance (#4133)
Observer007 Jul 27, 2026
d536efd
fix(gdn): support WY decode on SM121 (#4117)
kahyunnam Jul 27, 2026
e683e30
fix(norm): convert float2 to e4m3 directly in packed cast (#4167)
elwhyjay Jul 27, 2026
ec15f69
perf(test): bulk-precompile XQA decode kernels to cut test wall time …
bkryu Jul 28, 2026
cd1342a
[perf] Optimize TRT-LLM routing for high-expert, high-top-k workload…
jiahanc Jul 28, 2026
60783fb
feat(xqa): ragged Q and per-row sliding-window masking for speculativ…
yichengj0 Jul 28, 2026
417bbd2
test(jit): assert BMM export symlink under GEN_SRC_DIR (#4187)
kahyunnam Jul 28, 2026
8a94642
Feat/ulysses p2p a2a (#3820)
forrestl111 Jul 28, 2026
e63fc61
feat(moe_ep): MegaMoE framework integration ready: CUDA graph support…
mhoqueanik Jul 28, 2026
ba56578
docs: document CuTe prefill scheduling override (#4162)
kangbintNV Jul 28, 2026
da70a5c
docs: add missing trtllm_fp8_per_tensor_scale_routed_moe API entry (#…
kangbintNV Jul 28, 2026
f90e9c4
docs(mamba): document checkpointing varlen arguments (#4129)
hebo1221 Jul 28, 2026
e1eb406
Yanqinz/fix-gemm-and-grouped-mm-test-issue (#4185)
yanqinz2 Jul 28, 2026
2deed6c
feat(comm): extend trtllm_allreduce to SM12x and fix lamport buffer p…
yichengj0 Jul 28, 2026
bf77321
feat(moe): add unified unpacked pre-routed FP4 mode (#4104)
feih-nv Jul 28, 2026
4890932
feat(mla): support packed low-head and variable-Q decode (#4178)
PerkzZheng Jul 28, 2026
45728da
bump version to 0.6.16 (#4142)
jimmyzho Jul 28, 2026
843043b
[fix]fix xqa flaky test on spark (#4161)
qsang-nv Jul 28, 2026
4a3b075
fix: make mxfp8 gemm test pass by having it quantize along the correc…
jdebache Jul 28, 2026
1d97cbb
fix(moe): serialize CuTe DSL autotune replay (#4192)
zianglih Jul 28, 2026
b575db1
feat(moe_ep): fault-tolerance rank mask (NCCL-EP + NIXL-EP) (#4183)
Anerudhan Jul 28, 2026
6812ddb
fix(xqa): fix PDL load ordering and SM90 fp8 draft-mask dispatch (#4199)
yichengj0 Jul 28, 2026
1ca039e
feat(msa): accept K/V views split from a packed paged KV cache (#4039)
yichengj0 Jul 29, 2026
7adc546
feat(moe): add TRTLLM MXFP4 W4A8 and W4A16 unified API support (#4159)
feih-nv Jul 29, 2026
f6c2f1e
[cli] add CLI helper for flashinfer-jit-cache and flashinfer-cubin wh…
dierksen Jul 29, 2026
a34a735
[feat] Add SITU trtllmgen MOE (#4180)
jiahanc Jul 29, 2026
92274ba
feat(sm120): fused MoE (SwiGLU) via moe_gemm is_gated for cute SM120 …
CarstyYou Jul 29, 2026
8b6af1a
fix(moe): pad trtllm-gen route map by one element to avoid OOB read (…
syuoni Jul 29, 2026
d193c3f
perf(moe_ep): CuTe-DSL 4.5.2 mainloop WAR — drop the 4.6.1 runtime fl…
mhoqueanik Jul 29, 2026
57b7719
Fix the routing inconsistency for num_groups > 1 (#3946)
b8zhong Jul 29, 2026
43f12df
fix: support host global scale in CuTe-DSL NVFP4 quantization (#4138)
akurathiswaraj Jul 29, 2026
4b964ec
fix/test(moe_ep): self-bootstrap 1-rank process group in dg mega orac…
mhoqueanik Jul 30, 2026
a02d94d
feat: support native qk_rope_head_dim=0 sparse MLA decode in trtllm-g…
JustinTong0323 Jul 30, 2026
b1055a7
feat(topk): support separate page table row starts (#4169)
zianglih Jul 30, 2026
e37a777
feat(comm): make mixed-comm VMM workspaces checkpointable (#3910)
galletas1712 Jul 30, 2026
5b1af97
SM 107 Reland + Merge Back from v0.6.16 Release Branch (#4280)
Vinnie6167 Jul 30, 2026
369d1ac
test(sm103): fix FP4 autotuner cache inspection (#4145)
tiffany940107 Jul 30, 2026
3b67976
bump version to 0.6.17 (#4283)
aleozlx Jul 30, 2026
71e1745
feat(gemm): sync mm_fp4 SM120 NVFP4 dense GEMM kernel to b12x HEAD (#…
yichengj0 Jul 31, 2026
39f2ce4
fix(b12x): correct fp4 quantization numerics and add input_global_sca…
yichengj0 Jul 31, 2026
cba9994
feat(moe): support shared expert fusion for trtllm-gen fp4 moe (#4239)
Aneureka Jul 31, 2026
ac76e18
test(moe): add tests for trtllm-gen fused MoE with GeGLU activation (…
Aneureka Jul 31, 2026
dca5546
perf: optimize trtllm_fmha_v2 fp8 causal attention q-tile scheduling …
akhilg-nv Jul 31, 2026
668a1ba
docs: improve Ulysses communicator and MoE EP docs (#4240)
kangbintNV Jul 31, 2026
e9bb80f
fix(docker): stop pip from swapping the +cuXXX torch in CI image (#4284)
bkryu Jul 31, 2026
2e97a60
fix: skip LogitsTransform on lanes beyond the split-KV chunk boundary…
yichengj0 Jul 31, 2026
d020372
fix(test): repair main CI regressions from #4280 (artifacts Rubin pin…
bkryu Aug 1, 2026
08ddfbc
fix: preserve DeepSeek no-group sigmoid routing weights (#3875)
alexeldeib Aug 3, 2026
ed6f615
fix(gdn): use block-end decay for SM100 state updates (#4311)
guangyunh-nv Aug 3, 2026
1d2fc15
feat(cake_kda): add optimized B200 recurrent prefill backend (#4262)
yyihuang Aug 3, 2026
4433996
feat(cake_kda): add optimized B200 recurrent decode backend (#4279)
yyihuang Aug 3, 2026
76c5836
test: Add sharding support to scripts/task_run_unit_tests.sh (#4141)
righthandabacus Aug 3, 2026
da42be0
test(msa_ops): fix stale split-K heuristic expectation on high-SM GPU…
jimmyzho Aug 3, 2026
5a4b54b
perf(moe): sync SM12x NVFP4 fused-MoE kernels to b12x HEAD (#4285)
yichengj0 Aug 3, 2026
28ca04e
feat(moe): sync SM12x W4A16 fused MoE family to b12x HEAD (#4255)
yichengj0 Aug 3, 2026
5192059
fix: support fp8 e5m2 output in rmsnorm_quant and fused_add_rmsnorm_q…
elwhyjay Aug 3, 2026
ff34b88
feat(moe): add packed per-tensor FP8 and BF16 FromLogits routing to u…
feih-nv Aug 4, 2026
496cc1d
perf: remove dead cudaGetDeviceProperties in sm120 groupwise GEMM (#3…
aws-jiadingg Aug 4, 2026
6c18be5
fix(quantization): use 64-bit row addressing in per-token NVFP4 quant…
S1ro1 Aug 4, 2026
02ccd88
Fix duplicated words in API documentation (#4194)
cupkk Aug 4, 2026
d7e390c
feat(cake_kda): share recurrent prefill kernels across SM100 family (…
yyihuang Aug 4, 2026
6fb14ef
feat(moe): enable BiasType::Mn (LoRA delta) for nvfp4/mxfp4 MoE (#3987)
zetacat Aug 4, 2026
5116e2b
chore: drop unused <nvrtc.h> includes from three launchers (#4293)
aleozlx Aug 4, 2026
67f7637
perf: remove NVFP4 TMA input padding copy (#4210)
Barry-Delaney Aug 4, 2026
4967994
Fix the expert correction bias checking inconsistency in `trtllm_mxin…
b8zhong Aug 4, 2026
c9d66fe
Update CODEOWNERS for moe_ep and autotuner sections (#4332)
aleozlx Aug 4, 2026
68a85cd
feat(moe): enable MxInt4 in unified API (#4320)
feih-nv Aug 4, 2026
968fa04
feat(cake_kda): share recurrent decode kernels across SM100 family (#…
yyihuang Aug 4, 2026
0961af9
[chore] Add jiahanc to gemm code owner (#4327)
jiahanc Aug 4, 2026
180f0d6
feat(kda): add fused Kimi K3 decode kernel (#4243)
djmmoss Aug 5, 2026
339a8f4
feat(cake_tinygemm2): add CAKE-generated SM100/SM103 tinygemm2 varian…
xslingcn Aug 5, 2026
7dce283
feat: add sm100 cp delta rule prefill (#4078)
guangyunh-nv Aug 5, 2026
38bf507
fix(cake_kda): support non-aligned recurrent prefill head counts (#4351)
yyihuang Aug 5, 2026
6c85301
Revert "test: Add sharding support to scripts/task_run_unit_tests.sh"…
bkryu Aug 5, 2026
d9c97ac
perf(topk): skip output index sort for tie-break selection (#4295)
zianglih Aug 5, 2026
dc4d214
Upgrade CuTe DSL FMHA cubins (#4291)
xrq-phys Aug 5, 2026
aa81e71
fix(moe): handle CuTe DSL finalize output tails (#4186)
S1ro1 Aug 5, 2026
af3f5e4
MoE monokernel Bug fix, barrrier remove and kernel rewrite. (#4027)
yugong333 Aug 5, 2026
f8ca34a
MSA decode path improvements (#4324)
rmhaskarnvidia Aug 5, 2026
985302c
Add CuTe DSL HCA backend for DeepSeek V4 sparse MLA (#3943)
myu-guo Aug 5, 2026
b567143
Add @StudyingShao to CODEOWNERS for multiple sections (#4364)
aleozlx Aug 5, 2026
3c57ef1
fix(moe_ep): let the host framework pin the CUDA device via Bootstrap…
mhoqueanik Aug 5, 2026
4b6f2ea
[feat] Add Blackwell cutedsl BF16 splitk dense gemm (#4266)
jiahanc Aug 6, 2026
e493ed8
Accept unpacked pre-routed topk in fp8 block-scale and bf16 MoE (#4352)
b8zhong Aug 6, 2026
919866b
fix(comm): keep fd_exchange importable on Python 3.10/3.11 (#4354)
nv-guomingz Aug 6, 2026
307e2fe
feat: log git commit hash on import (FLASHINFER_LOGLEVEL >= 1) (#4286)
aleozlx Aug 6, 2026
3e96dfa
fix(gemm): gate cuDNN out of SM12x bmm_fp8 auto when override_shape u…
Saddss Aug 6, 2026
e2ea0e8
perf: normalize autotuner nearest-profile cache keys (#3984)
wongsingfo Aug 6, 2026
e44dae5
feat(moe): add CuTe-DSL GeGLU-tanh and SiTU activations (#4009)
murphymatt Aug 6, 2026
89a2592
bump version to 0.6.18 (#4384)
aleozlx Aug 6, 2026
c9172e6
feat(moe): support BF16 activations in CuTe DSL NVFP4 MoE (#4048)
zianglih Aug 7, 2026
0263dc2
docs: document tinygemm2 escape hatch and MonoMoE scratchpad (#4381)
kangbintNV Aug 7, 2026
553c228
feat: optimize gated SM12x dynamic NVFP4 MoE (#4329)
EricChen02 Aug 7, 2026
f2e0440
fix(monomoe): restore CUDA 12.0+ compatibility in tma_load_2d (#4406)
aleozlx Aug 7, 2026
b1d9585
feat(moe_ep): SM90 (Hopper) pull-style FP8 mega-MoE backend (#4113)
mhoqueanik Aug 8, 2026
6d4f309
feat(topk): Add top_k_varlen with GVR and radix backends for sparse-a…
dhiraj113 Aug 8, 2026
24d7dfb
feat(sm120): consolidate DSV4 sparse MLA top-k 192/256 support (#4380)
lucifer1004 Aug 8, 2026
29add4e
Fix/quantized allreduce ci harness (#4330)
cindyzxq Aug 8, 2026
29196cf
perf(moe): persist b12x MoE CuTe-DSL kernels to the disk cache (#4331)
Smallfu666 Aug 8, 2026
7f614b8
feat(topk): support compact page table transforms (#4315)
zianglih Aug 9, 2026
4fbac49
fix(moe_ep): preserve singleton expert TMA modes (#4296)
cherichy Aug 9, 2026
aadaae7
feat(moe): add unified CUTLASS BF16 and W4A16 runners (#4328)
feih-nv Aug 10, 2026
2ab910c
feat(sm120): add native MXFP4 W4A4 fused MoE (#4290)
Yuening-wa Aug 10, 2026
2fb785c
fix(moe): pad BF16 TRTLLM-Gen intermediates to 128 KiB (#4319)
nvpohanh Aug 10, 2026
a01a52e
require explicit ci triggering for all pull requests (#4383)
yongwww Aug 10, 2026
efa05fb
test incidental support for fp32 unpacked expert weights from #3763 (…
jdebache Aug 11, 2026
e31f4dd
fix(moe): prepare MXFP8 MXFP4 profiler inputs (#4308)
ormandj Aug 11, 2026
ef4e6d1
Feat/deepseek fused moe fp8 blockwise swigluoai (#4405)
vitamin-chaos Aug 11, 2026
906181e
feat(comm): add Blackwell MNNVL CuTe DSL all-reduce fusion backend (#…
qiangyicheng Aug 11, 2026
42ea835
perf(gdn): optimize cp host launch overhead for sm90 and sm120 (#4374)
guangyunh-nv Aug 11, 2026
b1a8782
fix(moe): use per-expert Humming residual scales in SM90 CUTLASS MXFP…
StudyingShao Aug 11, 2026
f527314
perf(norm): speed up fused add RMSNorm FP4 quantization (#4416)
soodoshll Aug 11, 2026
e673f7f
Add task-scheduled Blackwell attention kernels (#4357)
PerkzZheng Aug 11, 2026
7af3100
fix(gemm): validate cached CuTeDSL BF16 tactics (#4455)
jiahanc Aug 11, 2026
78e463b
perf(cake_tinygemm2): add a STAGES=16 kernel tier for single-wave lar…
xslingcn Aug 11, 2026
e9fb62b
perf(moe): optimize CUTLASS fused MoE mem-bound kernels (#3761)
Aneureka Aug 11, 2026
db9293d
remove spot instances from pr-test, use on-demand only (#4454)
yongwww Aug 11, 2026
1b92281
test: Add sharding support to scripts/task_run_unit_tests.sh (#4359)
righthandabacus Aug 11, 2026
d57bfb1
feat(gdn): u/d cache spec-decode kernels for replayssm (#4081)
ameynaik-hub Aug 11, 2026
bac0eb7
Mirror nightly release to flashinfer-ai/whl repo (#4353)
dierksen Aug 12, 2026
0ef086a
perf(sm120): wave+residue tile-selection for plain groupwise MoE GEMM…
CarstyYou Aug 12, 2026
0659712
refactor(moe): enforce build() for all runners and refine CUTLASS aut…
feih-nv Aug 12, 2026
c285307
Update Docker CI tags to 20260811-e673f7f (#4457)
flashinfer-bot Aug 12, 2026
fb5a69a
fix(gdn): convert fp16 decode operands and fix cache identity (#4219)
kahyunnam Aug 12, 2026
aaf97df
fix(fmha): propagate LSE reduction launch errors (#4389)
leejnau Aug 12, 2026
147739b
feat: paged KV cache support for the modular CuTe-DSL Blackwell prefi…
pgera Aug 12, 2026
9ffd995
chore: add KDA CODEOWNERS (#4482)
kahyunnam Aug 12, 2026
2d132e6
Fix HCA Gather4 support for public CuTe DSL (#4368)
myu-guo Aug 12, 2026
f9b13ef
feat(moe_ep): add SM90 push FP8 mega-MoE backend for Hopper (#4069)
leonardHONG Aug 12, 2026
b8c2192
fix(comm): align MNNVL two-shot workspace stages (#4473)
aoshen02 Aug 12, 2026
27b8a80
Fused shared experts support via the unified API (#4326)
feih-nv Aug 13, 2026
8f9ad20
feat(attention): asymmetric VO-split NVFP4 paged prefill (qk=512/vo=2…
jethac Aug 13, 2026
fcf2604
ci: centralize minimum Python version at 3.10 (#4464)
dierksen Aug 13, 2026
53a1c3b
perf(gemm): split-K, occupancy, and m=1 stream-GEMV decode tactics fo…
yichengj0 Aug 13, 2026
0d25b18
feat(moe): MXFP8 x MXFP4 CuTe-DSL fused MoE for SM100, plus large-bat…
vitamin-chaos Aug 13, 2026
b065838
Add distribution-aware autotuning for TRT-LLM MoE (#4106)
samuel21119 Aug 13, 2026
c517c07
feat(quantization): cuTile per-token-group 8bit quant + fused RoPE-FP…
yifeis-nv Aug 13, 2026
2febce5
feat(gemm): port SM12x MXFP8 dense GEMM from b12x (#4305)
yichengj0 Aug 13, 2026
ba6bf4e
perf(gdn): reuse pretranspose kernels across pool capacity and stride…
leonardHONG Aug 13, 2026
ed4ed65
chore(codeowners): add @jimmyzho to tests/attention/ (#4508)
jimmyzho Aug 13, 2026
140be0e
skip the deliberate-trap test under compute-sanitizer (#4493)
qsang-nv Aug 13, 2026
d69ab74
feat(msa): fp8 index-k and index-q support in the SM12x proxy-score k…
yichengj0 Aug 13, 2026
d898ed0
Fix TRTLLM ragged prefill edge cases (#3779)
alexeldeib Aug 13, 2026
1711993
Support non-interleaved KV input to fmha_v2 (#4124)
akhilg-nv Aug 13, 2026
61011f7
feat(attention): cuDNN paged prefill via direct mixed-form cu_seq_len…
egilliam-nv Aug 13, 2026
ed6c709
perf(gdn): reuse MTP decode kernels across cache modes (#4128)
hebo1221 Aug 13, 2026
5931d44
ci: add PR API and documentation checks (#3917)
kangbintNV Aug 14, 2026
5fc667f
Add SM120 FP8 FMHAv2 self-attention (#4272)
rosenrodt Aug 14, 2026
a9e03bf
feat(kda): add packed-input CuTe decode kernel (#4417)
ameynaik-hub Aug 14, 2026
555492e
chore: explicitly set mMultiCtasKvMode in ragged attention launcher (…
namgyu-youn Aug 14, 2026
750dbfd
integrate SM120 VSA (Video Sparse Attention) block-sparse backend (#…
hsr1234563 Aug 14, 2026
23382fd
ci: upgrade sccache and retain stats (#4505)
dierksen Aug 14, 2026
4fa40ae
fix: correct trtllm-gen reduction indexing for FMHA decode at long q_…
jhaotingc Aug 14, 2026
4757ebc
ci: upload GitHub unit-test JUnit reports (#4488)
dierksen Aug 14, 2026
3f5acea
perf(moe): fold the SiTU beta reciprocal in the CuTe DSL MoE epilogue…
sychen52 Aug 14, 2026
9df0de2
feat: unpacked FP8 per-tensor scaling support for TRTLLM fused MoE (#…
jdebache Aug 14, 2026
d1fff56
refactor(moe_ep): taxonomy/provenance restructure; incorporate SM90 p…
mhoqueanik Aug 14, 2026
a6dfe70
gemm: enable CUTLASS NVFP4 SVDQuant on SM107 (#4509)
Vinnie6167 Aug 14, 2026
956c537
ci: grant @flashinfer-bot access to collaborators who can apply label…
yongwww Aug 14, 2026
38ebb15
feat: support FP8 KV NoPE MLA on SM90 (#4373)
JustinTong0323 Aug 14, 2026
0456774
fix(ci): correct documentation finding locations and skipped runs (#4…
kangbintNV Aug 15, 2026
40bb691
test(gdn): trim GDN decode test matrix to distinct kernel specializat…
kahyunnam Aug 15, 2026
8044d94
feat: add BF16Q FP8KV transform-mode selection (#3544)
djmmoss Aug 15, 2026
8ac4558
[prims-ts] Refactor FMHA decode padding-task scheduling (#4413)
IwakuraRein Aug 16, 2026
f4043ef
feat(moe): support GeGLU and SiTU in W4A16 (#4394)
zianglih Aug 16, 2026
e77a4a0
feat(cake_kda): add strided prefill state checkpoints and packed deco…
yyihuang Aug 16, 2026
01e3179
WIP jit-cache wheel size fix: drop 12.1a from the cu129 aarch64 jit-c…
kahyunnam Aug 17, 2026
a91f9fa
docs: include top_k in topk API docs (#4501)
kangbintNV Aug 17, 2026
9cd1fcf
Fix/missing source tree files (#4517)
cindyzxq Aug 17, 2026
d9463fb
ci: install quack-kernels with --no-deps so it cannot downgrade CuTe …
kahyunnam Aug 17, 2026
9a7c67d
test: make heavy attention test files more efficient (bulk JIT precom…
bkryu Aug 17, 2026
829f03a
fix(moe): release W4A16 CuTe DSL scratch workspaces (#4558)
zianglih Aug 18, 2026
02647e8
[fix] disable tileN=192 for trtllmgen per-token NVFP4 MoE (#4563)
jiahanc Aug 18, 2026
94f6ae8
feat: sm100 cute_dsl w4a16 gemm (#4466)
IwakuraRein Aug 18, 2026
a25969f
feat(b12x): support MXFP4 dense GEMM on SM120 (#4479)
ayrnb Aug 18, 2026
0751be9
feat: Grouped-token MLA support for the TRTLLM-Gen FMHA backend. (#3849)
farazkh80 Aug 18, 2026
27a5a29
fix(build): restore the nixl v1.3.1 submodule pin accidentally rolled…
mhoqueanik Aug 18, 2026
4581d63
Fix/missing source tree files for two more moe_ep test files (#4580)
cindyzxq Aug 18, 2026
065795e
fix(xqa): out-of-bounds read of attention sinks in SM90 kernel epilog…
bkryu Aug 18, 2026
efbd232
feat(moe): add TopKSigmoid routing method (TopK -> Sigmoid) (#4404)
EdalatiAli Aug 18, 2026
fafab47
feat(xqa): support non-MLA head_dim 512 (Gemma-style GQA decode) on S…
bkryu Aug 18, 2026
7319c44
perf(moe): fold W4A16 SiTU reciprocals (#4540)
zianglih Aug 18, 2026
92ef090
fix: guard oversized shared memory driver enums with CUDA >= 13.4 (#4…
JiaxuanBai Aug 18, 2026
7015afd
[MoE] Expose zero-copy MegaMoE workspace output view (#4341)
foraxe Aug 18, 2026
7aa0cd3
refactor(moe): split finalize knobs out of ExecutionConfig into MoEFi…
aleozlx Aug 18, 2026
4923189
feat(cake_kda): add optimized H12 packed decode across SM100 family (…
yyihuang Aug 18, 2026
d417f44
ci: use the CI image's packages and share a pip cache between jobs (#…
yongwww Aug 19, 2026
61a6c65
stabilize pytest node IDs for external scopes to fix issues 4499 and …
righthandabacus Aug 19, 2026
d5e45b3
feat: CuTe DSL kernels for Rubin (SM107) and batched FP8 GEMM for Bla…
kahyunnam Aug 19, 2026
87ba7ff
create rc4
kahyunnam Aug 19, 2026
c3b96d0
ci: build jit-cache wheels for CUDA 13.4 with 10.7a (Rubin/SM107)
Vinnie6167 Aug 18, 2026
0e9c946
fix(trace): model the SM100/103 cute-dsl mm_bf16_fp4 prepared layout
kahyunnam Aug 19, 2026
57c6951
fix(moe): restore SM12x MoE kernels broken by self-resolved helper in…
lucifer1004 Aug 19, 2026
549a7ce
add backward-compatible aliases for bsa_attn_fwd and bsa_attn_blk64_f…
hsr1234563 Aug 19, 2026
a35db6c
revert: #3975 and #4129 (mamba checkpointing ssu 2-split + doc)
kahyunnam Aug 19, 2026
66b1a12
create rc5
kahyunnam Aug 19, 2026
84c1aa5
test: enable the unified MoE fuzzer by default and prune legacy UTs (…
feih-nv Aug 19, 2026
de96e24
create rc6
kahyunnam Aug 20, 2026
211c189
fix(gemm,moe): restore bmm_fp8 auto fallback and drop an over-strict …
Vinnie6167 Aug 21, 2026
1629aba
feat: trtllm-gen FMHA features for sm107 (spcompress, fp16softmax) (#…
jimmyzho Aug 20, 2026
76f8fe6
fix(moe): unify the NVFP4 profiler workspace predicate and fix gated …
yichengj0 Jul 16, 2026
34f84cb
feat: collect a union of TEST_PATH targets in unit CI (#4641)
kahyunnam Aug 20, 2026
3b3e688
create rc7
kahyunnam Aug 21, 2026
6ad4655
fix(gdn): support WY ucache kernels on SM121
hebo1221 Aug 15, 2026
98b0a6d
fix(attention): enable SM121 FMHA v2 prefill (#4661)
aeichler-ac Aug 25, 2026
24f9b90
fix(cute_dsl): consult the arch gate in the GEMM and GDN dispatchers …
Vinnie6167 Aug 25, 2026
325096f
fix: support SM107 (Rubin) in topk_varlen, FlashKDA, and router GEMM …
Vinnie6167 Aug 25, 2026
23381da
fix: align the nvidia-cutlass-dsl floor on release-v0.6.18 with main …
kahyunnam Aug 25, 2026
49a47af
create rc8
kahyunnam Aug 25, 2026
2852cb4
ci: pin nvidia-cutlass-dsl 4.6.2 on release-v0.6.18 (#4758)
kahyunnam Aug 26, 2026
c46060f
ci: keep Rubin cute-dsl pins after the 4.6.2 CI pin (#4762)
kahyunnam Aug 26, 2026
21a95d7
fix(kda): fall back to Cake when CuTe DSL predates cutlass.experiment…
kahyunnam Aug 26, 2026
dc642a9
fix(sparse-mla): prevent SM120/SM121 prefill KV-handshake hang (#4732)
kahyunnam Aug 26, 2026
f703c8e
Revert "ci: build jit-cache wheels for CUDA 13.4 with 10.7a (Rubin/SM…
kahyunnam Aug 26, 2026
6677925
ci: consolidate CUDA coverage and validate candidate images (#4469)
dierksen Aug 22, 2026
d23e7d4
ci: disable sccache for cu134 nvcc (#4682)
dierksen Aug 24, 2026
8edb1f3
ci: coordinate CUDA dependency policy (#4711)
dierksen Aug 25, 2026
117883f
fix(ci): skip source-only CUDA config test in nightlies (#4750)
dierksen Aug 26, 2026
89aa0fe
build(jit): reduce JIT-cache wheel size (#4757)
kahyunnam Aug 26, 2026
c8e9e59
chore(aot): exclude single prefill/decode modules from jit-cache preb…
kahyunnam Aug 26, 2026
7f17ca2
feat(prims_ts): accept SM107 (Rubin) in the attention-ts device gate
Vinnie6167 Aug 26, 2026
066a52f
fix: SM107 (Rubin) MoE tactic guard and batched-GEMM cubin pin (#4761)
kahyunnam Aug 26, 2026
1073859
fix(cute_dsl): make the optional-dependency guard independent of cutl…
kahyunnam Aug 26, 2026
c2becae
Refresh trtllm-gen cubins after multi-arch generation is supported
jimmyzho Aug 20, 2026
3c17a4f
ci: install cutlass-dsl 4.6.2 with its libs wheels
kahyunnam Aug 27, 2026
59b16e6
fix(cute_dsl): restore the GDN compile arch helper
kahyunnam Aug 27, 2026
eadfae3
create rc9
kahyunnam Aug 27, 2026
dfdc50a
fix: SageAttention support block size doesn't divide sequence; suppor…
xrq-phys Aug 25, 2026
9556255
fix: skip SM107 low-latency GEMM cubins on Blackwell (#4773) (#4786)
kahyunnam Aug 27, 2026
9ac0f5d
fix(moe): filter the trtllm-gen kernel manifest per arch to restore M…
jimmyzho Aug 27, 2026
89cafe9
[release-v0.6.18] Rubin (SM107) open-issue fixes (#4790)
Vinnie6167 Aug 27, 2026
f0ff7f3
[release-v0.6.18] Five SM107 (Rubin) fixes: 133 CI failures (#4792)
Vinnie6167 Aug 28, 2026
e62941a
create rc10
kahyunnam Aug 28, 2026
ef04937
fix(moe): disable PDL in the trtllm-gen fused-MoE path on SM107
Vinnie6167 Aug 28, 2026
69ff11f
bump version to 0.6.18
kahyunnam Aug 28, 2026
34f8f0e
chore: record v0.6.18 as an ancestor of the fork
demandal25 Sep 2, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view

These merge commits were added into this branch cleanly.

There are no new changes to show.