Void python3-pywhispercpp CPU backend is built without AVX/FMA/F16C #44

New Issue

2026-11-06T17:26:53Z

xavierk commented

2026-11-06 17:26:53 +00:00

Found in the hardware acceptance run (#19). CPU real-time factor is 1.3-1.7 on the 30 s sample because libggml-cpu is built portable without AVX, FMA or F16C (kept portable on purpose so the package runs on any x86_64). Decide: ship ggml's runtime CPU-variant loading (GGML_CPU_ALL_VARIANTS / dynamic backends) instead of a single baseline build. Evidence: #19 (comment)

Found in the hardware acceptance run (#19). CPU real-time factor is 1.3-1.7 on the 30 s sample because libggml-cpu is built portable without AVX, FMA or F16C (kept portable on purpose so the package runs on any x86_64). Decide: ship ggml's runtime CPU-variant loading (GGML_CPU_ALL_VARIANTS / dynamic backends) instead of a single baseline build. Evidence: https://git.bongbetic.com/xavierk/voculinux/issues/19#issuecomment-12913

xavierk added the needs-triage label 2026-11-06 17:26:53 +00:00

xavierk referenced a pull request that will close this issue

2026-11-06 18:49:32 +00:00

fix(void): report pywhispercpp 1.5.0 and load CPU variants at runtime #49

xavierk referenced this issue from a commit

2026-11-06 18:49:32 +00:00

fix(void): report pywhispercpp 1.5.0 and load CPU variants at runtime

xavierk closed this issue

2026-11-06 18:51:30 +00:00

xavierk referenced this issue from a commit

2026-11-06 18:51:32 +00:00

fix(void): report pywhispercpp 1.5.0 and load CPU variants at runtime (#49)