ci: windows-directml integration job on self-hosted Windows GPU runners (#292) - #296

Merged

beveradb merged 11 commits into

mainfrom

feat/sess-20260720-1450-windows-directml-job

Jul 20, 2026

Commits on Jul 20, 2026

ci: add windows-directml integration job on self-hosted Windows GPU runners (#292)

beveradb
and
claude
committed
7048cab
fix(ci): avoid Windows MAX_PATH failure in windows-directml poetry install

beveradb
and
claude
committed
a5e99a9
fix(ci): install VC++ runtime for windows-directml job

beveradb
and
claude
committed
44f5c45
fix: einsum attention fallback on DirectML + CLI exits nonzero on failed separation

beveradb
and
claude
committed
d72436c
fix: compute rotary embeddings without zero-width concat on DirectML

beveradb
and
claude
committed
3a70443
fix: bound attention memory on DirectML with batch-sliced einsum

beveradb
and
claude
committed
1462b41
fix: use matmul (not einsum) for DirectML attention; add MDX23C to DML matrix

beveradb
and
claude
committed
00d7222
test: reduced MDXC segment size on DirectML (allocator can't sustain 801 fp32)

beveradb
and
claude
committed
3cc43ae
fix: release torch-directml allocator memory between demix batches

beveradb
and
claude
committed
6dd57c7
fix: MDXC/RoFormer models run on CPU under DirectML (allocator limitation)

beveradb
and
claude
committed
530f65d
test: assert new-implementation load via separator stats line; widen DML job timeout

beveradb
and
claude
committed
c0215f9