Skip to content

[TRTLLM-15917][feat] Integrate Sol-Attn sparse attention into VisualGen #18329

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
chang-l merged 22 commits into from
Sep 19, 2026
Merged

[TRTLLM-15917][feat] Integrate Sol-Attn sparse attention into VisualGen #18329

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
22 commits
Select commit Hold shift + click to select a range
73a0b94
[TRTLLM-15917][feat] Integrate Sol-Attn sparse attention into VisualGen
karljang Aug 28, 2026
c003e55
[TRTLLM-15917][fix] Keep the Sol-Attn CuTe DSL launch opaque to Dynamo
karljang Sep 1, 2026
47c7cbf
[TRTLLM-15917][doc] Declare sol_attn_backend.py's upstream derivation
karljang Sep 1, 2026
5e077c1
[TRTLLM-15917][feat] Drop sm120 from the Sol-Attn port
karljang Sep 2, 2026
8593ad8
[TRTLLM-15917][fix] Keep Sol-Attn's non-sparse attention on the confi…
karljang Sep 2, 2026
bd47daa
[TRTLLM-15917][fix] Restrict the in-family fallback to true cross-att…
karljang Sep 3, 2026
d3eb260
[TRTLLM-15917][chore] Rename SolAttnAttention to SolAttention
karljang Sep 3, 2026
1f55662
[TRTLLM-15917][refactor] Decide Sol-Attn eligibility per call, not pe…
karljang Sep 3, 2026
a37e7ff
[TRTLLM-15917][fix] Address review findings on the Sol-Attn integration
karljang Sep 3, 2026
c784739
[TRTLLM-15917][feat] Support SM103 (B300/GB300) for Sol-Attn
karljang Sep 8, 2026
da5a5a4
[TRTLLM-15917][fix] Address review feedback on Sol-Attn
karljang Sep 9, 2026
a2c978f
[TRTLLM-15917][fix] Reuse SkipSoftmaxScheduler for the Sol-Attn graph…
karljang Sep 9, 2026
32e9ab9
[TRTLLM-15917][test] Make ineligibility-reason cases reach the branch…
karljang Sep 10, 2026
3de388f
[TRTLLM-15917][fix] Honor masks, survive CUDA-graph capture, wire LTX…
karljang Sep 11, 2026
2a2365b
[TRTLLM-15917][chore] Drop an unused variable left in the multi-block…
karljang Sep 11, 2026
ae7c894
[TRTLLM-15917][test] Teach the LTX-2 two-stage test stubs the CUDA-gr…
karljang Sep 11, 2026
09bb2c7
[TRTLLM-15917][fix] Keep the CUDA-graph phase scope in LTX-2's captur…
karljang Sep 14, 2026
cd8ded2
[TRTLLM-15917][chore] Drop the no-op kv_splits option from SolAttenti…
karljang Sep 14, 2026
205c8d5
[TRTLLM-15917][chore] Vendor the Sol-Attn kernel through vendor_sources
karljang Sep 16, 2026
20b1424
[TRTLLM-15917][fix] Address review: typed dense_layers, TRTLLM_ stric…
karljang Sep 16, 2026
bf3652a
[TRTLLM-15917][chore] Clarify dense-backend initialization rationale
karljang Sep 17, 2026
b901f29
[TRTLLM-15917][chore] Align the SEPARATE_QKV fallback with main's cro…
karljang Sep 18, 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
2 changes: 1 addition & 1 deletion .pre-commit-config.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1525,7 +1525,7 @@ legacy-files: &legacy_files |
# list; the hook's own `files:` pattern only gates *when* the hook triggers.

# Global exclude: vendored code + trtllm-gen FMHA artifacts (cubin pointers, export headers, cuda_ptx)
exclude: '(^cpp/tensorrt_llm/common/sha256/|^triton_kernels/|^tensorrt_llm/_torch/attention/backends/prims_ts/|trtllmGenKernels/fmha/cubin/kernelMetaInfo\.h$|cubin\.cpp$|cubin\.h$|trtllmGenKernels/fmha/trtllmGen_fmha_export/|trtllmGenKernels/fmha/cuda_ptx/|trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelMetaInfo\.h$|trtllmGenKernels/gemm/trtllmGen_gemm_export/KernelMetaInfo\.h$|\.cubin\.tar\.zst$)'
exclude: '(^cpp/tensorrt_llm/common/sha256/|^triton_kernels/|^tensorrt_llm/_torch/attention/backends/prims_ts/|^tensorrt_llm/_torch/visual_gen/cute_dsl_kernels/blackwell/sol_attn/|trtllmGenKernels/fmha/cubin/kernelMetaInfo\.h$|cubin\.cpp$|cubin\.h$|trtllmGenKernels/fmha/trtllmGen_fmha_export/|trtllmGenKernels/fmha/cuda_ptx/|trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelMetaInfo\.h$|trtllmGenKernels/gemm/trtllmGen_gemm_export/KernelMetaInfo\.h$|\.cubin\.tar\.zst$)'

default_install_hook_types: [pre-commit, commit-msg]
repos:
Expand Down
Loading
Loading