feat: CUDA IPC zero-copy GPU transport (TD↔SD input, output, ControlNet)#15

Open

forkni wants to merge 24 commits into

SDTD_031_devfrom

feat/cuda-ipc-output

Commits on May 16, 2026

chore: gitignore local cgw tooling; add PR audit doc
forkni
committed
feat: add FP8 QDQ finite-scale gate and fused-MHA layer count
forkni
committed
feat: add quality regression harness with FP16-TRT goldens and SSIM/LPIPS metrics
forkni
committed
feat: port varshith15 kvo_cache patch onto diffusers 0.38.0 via runtime monkey-patch
forkni
committed
fix: replace emoji chars in SDXL ONNX size warning to avoid cp1252 encode error on Windows
forkni
committed
feat: seed quality-harness goldens, manifest, thresholds; fix FP8 CFG batch mismatch in calibration
forkni
committed
chore: stage pre-existing formatter diffs (quote/whitespace normalisation)
forkni
committed
fix: per-input-aware FP8 calibration tile for static-dim0 ONNX inputs

forkni
and
claude
committed

Commits on May 17, 2026

fix: kvo_cache patch breaks ControlNet ONNX export — sentinel for backward-compat return arity
forkni
committed

Commits on May 18, 2026

Commits on May 23, 2026