[hannahli-nv]

Description

Update codes.

This PR contains 18 new commit(s).

Commits included:

c3e4c0a fix(dropout): use int64 offsets in the cuTile kernel
03b2778 docs(roadmap): update the suite status table
5d24c26 docs(license): add flash-linear-attention and Liger-Kernel attributions
7c63216 fix(benchmark): pace timed GPU work below the power-capping window
a34b9bf feat(softmax): add cuTile online softmax and softmax backward
29cd674 feat(fla): add cuTile linear-attention operators (chunk, fused-chunk, Mamba-2, DeltaNet, GDN)
c08bbb6 test(liger): patch the chunk thresholds of the backend under test in test_op_backward_chunked
f0e0f2c fix: drop skinny-GEMM aarch64 xfails
1e34b4c fix(tests): preserve complex components in tensor comparisons
89f25fb Preserve pointer alignment for Rust matmul kernels
95c6da5 chore(triton): tidy package comments
3ec2972 refactor(tilecpp): factor cubin compilation out of get_kernel
45c6e74 docs(wiki): cover GLU and block-scaled GEMM kernels
cde2456 docs(wiki): correct indexing, sigmoid, and lint guidance
d0a327d chore: tidy a kernel docstring
f206237 test(fmha): cool down before SOTA benchmarks
78f7e64 Update Julia kernels and skill to cuTile.jl 1.0
a7b7605 chore(hf-bench): upgrade transformers 5.10.2 -> 5.19.0 (+ tokenizers 0.23.3)

CI Configuration

config:
  build: true
  # valid options are "ops" and "benchmark"
  test: []

Checklist

• Code formatted and imports sorted via repo specifications (./format.sh)
• Documentation updated (if needed)
• CI configuration reviewed

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.