Skip to content

sa3.cpp v0.2.2

Pre-release
Pre-release

Choose a tag to compare

@betweentwomidnights betweentwomidnights released this 08 Nov 21:37
fabb001

sa3.cpp v0.2.2

Pre-release for gary4local validation. It becomes a full release once that passes.

Fixed

  • Negative prompts under CFG. stable_audio_3 gives the negative prompt's masked tokens the null embedding in the unconditioned pass; sa3.cpp gave them the padding embedding, so with a short negative prompt like "low quality" nearly the whole unconditioned context was wrong. Those rows are now zeroed, only when a negative prompt is given. Renders with cfg_scale != 1 and a negative prompt sound different; at cfg 1, or without a negative prompt, nothing changes. Against the PyTorch reference, the final latent's error drops from 3.4% to 0.24%. (#91)
  • Linux builds. sa3-lib-sat-smoke now links libm on non-Apple Unix, so ./build.sh completes on glibc. (#91)

Added

  • lora_lanes capability. /health.capabilities and --control-info report "lora_lanes": true, so a host can tell a lanes-aware server from one that would silently ignore automation. gary4juce's remote SA3 reports the same flag. (#91)

Changed (since v0.2.0)

  • ggml moves to upstream GGML 0.29. The submodule pins e3db5210 (ggml#22), up from b77ae790 (0.27). The fork's training kernels and backward ops carry over. (#90, first in the v0.2.1 pre-release)
  • CUDA and Vulkan renders change; Metal's don't. On CUDA, the same seed gives a slightly different take: one upstream commit sums RMS norm in a different order, so it is rounding, not a different computation. In a blind A/B no difference was audible. On Metal, every render and training output is byte-identical to v0.2.0.
  • Plugin builds trim source paths with SA3_PRIVATE_GGML (#89). This is off for these release packages, so they are unaffected.
  • The Windows CUDA backend is built once per ggml pin (#88). This release packages the ggml-cuda.dll that dry run 42518070848 compiled and attested; BUILD-INFO.json names that run.

The C ABI is unchanged. Packaging is the same as v0.2.0: Windows has the split core/CUDA/Vulkan and standalone ZIPs, and macOS the signed, notarized universal package.