SolitaryThinker
32d8a8765b
FastH3's tile-128 route runs the ptx_blackwell CUDA forward on sm_100/sm_103 and, on every other GPU,
splits each 128-token tile into two 64-token halves on the Triton kernels. On the DGX Spark that split
route is what OmniRef's BF16 attention runs today. This adds a 128-token forward built for sm_121a
(mma.sync + TMA; the GB10 has no tcgen05 or tensor memory) and routes the tile-128 call to it on a
GB10, ahead of the split.
- csrc/attention/block_sparse_kernel_blk128_sm121.cuh and block_sparse_blk128_sm121.cu: the kernel and
its binding block_sparse_sm121_blk128_fwd(q, k, v, q2k_idx, q2k_num, variable_block_sizes) -> {out},
inference only, no LSE.
- CMakeLists.txt: built when TORCH_CUDA_ARCH_LIST has 12.1a or 12.1 and CUDA >= 13, so both the
aarch64 wheel and the GB10 Docker image carry it.
- fastvideo_kernel.block_sparse_attn_sm121: is_supported and an opaque custom op with a fake. It is a
separate module because block_sparse_attn_ptx_blackwell.is_supported also gates an LSE-requiring
autograd path.
- video_sparse_attn_h3: the tile-128 route tries it after ptx_blackwell and before split_tile128.
- Tests: kernel correctness against an FP32 reference (ragged blocks, empty selections, odd block
counts, many waves, a NaN-filled allocator, determinism), and the H3 route (fakes, plus a real-GB10
oracle test).
FastVideo documentation
Canonical guides live in this directory. The Blume application in docs-site builds and publishes them at haoailab.com/FastVideo.
See Documenting FastVideo for local setup,
source ownership, page structure, validation, and deployment. The existing
GitHub Pages workflow publishes Blume from main with the same domain and base
path. MkDocs is no longer part of the build.
Cookbook sources
Recipe metadata, shared JavaScript, and tests remain under docs/. From the
repository root:
npm ci --prefix docs
npm run build:catalog --prefix docs
npm run check:cookbook --prefix docs
npm run test:cookbook --prefix docs
python -m unittest discover -s docs/tests -p 'test_*.py'
The combined browser bundle, notices, index and catalogs are generated together
under assets/cookbook-config/. Do not hand-edit or commit those outputs.
The site preparation builds and copies them into Blume's public assets.
See the recipe authoring guide and
configuration API contract before changing the schema.