FlashAttention#13

Merged

reeselevine merged 45 commits intomasterfrom

Jan 4, 2026

Commits on Oct 1, 2025

neg f16xf32xip builds and runs, havent actually ran a model that uses neg kernel yet though
XXjcontiniXX
committed
neg passes backend test
XXjcontiniXX
committed

Commits on Oct 9, 2025

unary operators pass ggml tests
XXjcontiniXX
committed

Commits on Oct 12, 2025

Commits on Oct 15, 2025

responded and dealt with PR comments
XXjcontiniXX
committed

Commits on Oct 30, 2025

implemented REPL_Template support and removed bug in unary operators kernel
XXjcontiniXX
committed
formatted embed wgsl and ggml-webgpu.cpp
XXjcontiniXX
committed

Commits on Nov 5, 2025

Faster tensors (#8)
reeselevine
authored

Commits on Nov 8, 2025

Use map for shader replacements instead of pair of strings
reeselevine
committed

Commits on Nov 11, 2025

Wasm (#9)

reeselevine
and
ngxson
authored

Commits on Nov 12, 2025

Commits on Nov 14, 2025

Commits on Nov 17, 2025

Disable multiple threads for emscripten single-thread builds in ggml_graph_plan
reeselevine
committed

Commits on Nov 18, 2025

Commits on Dec 11, 2025

Start work on flash attention
reeselevine
committed

Commits on Dec 16, 2025

Shader structure set up (many bugs still)
reeselevine
committed

Commits on Dec 17, 2025

Commits on Dec 19, 2025

Generalize softmax to work with multiple subgroups, f16 accumulation, mask shared memory tiling
reeselevine
committed

Commits on Dec 29, 2025

Start work on integrating pre-wgsl
reeselevine
committed

Commits on Dec 30, 2025

Commits on Dec 31, 2025

Commits on Jan 1, 2026

Commits on Jan 3, 2026

Commits on Jan 4, 2026

formatting
reeselevine
committed