Tencent-Hunyuan / FlashTopK

Fast, exact fused top-k selection for learned sparse attention training and inference, from HyScale (Tencent Hy & University of Edinburgh).

Why this job: matched by what the repo talks about: inference, fast, train, exact

1 of 2

What it does

Fast. the fastest open-source implementation.
Robust. stays fast on trained and untrained indexers. Adaptive key pruning decides per call whether to skip keys by a score bound.
Exact. the true top-k. Rows the sampled threshold misses are rescored on the GPU.
Deterministic. the same bits on every run. Ties at the k-th score keep the smallest key indices.
CP-compatible. each row's threshold comes from a strided sample of its own window, so context parallelism needs no selection from another rank.

pip install -e .
# torch >= 2.6, triton >= 3.3, nvidia-cutlass-dsl >= 4.5, apache-tvm-ffi
pip install -e .[test]
# pytest for the test suite

Repos similar to this one
Repo	Stars	Last commit	Licence	Downloads for
liguodongiot / llm-action	25,274	29 days ago	Apache-2.0	No downloads
little51 / llm-dev	238	8 months ago	MIT	No downloads

More: languages, topics and project record