From 0fd063442a0de371d2c8eed7e15cc8dcfc95dae5 Mon Sep 17 00:00:00 2001 From: Jamie Pine Date: Thu, 29 Jan 2026 21:08:28 -0800 Subject: [PATCH] Remove risks and mitigations section from MLX_AUDIO.md and update open questions with responses for clarity. This streamlines the documentation and provides clearer guidance on future considerations. --- docs/plans/MLX_AUDIO.md | 23 +++-------------------- 1 file changed, 3 insertions(+), 20 deletions(-) diff --git a/docs/plans/MLX_AUDIO.md b/docs/plans/MLX_AUDIO.md index d333b72b..c42f4064 100644 --- a/docs/plans/MLX_AUDIO.md +++ b/docs/plans/MLX_AUDIO.md @@ -280,28 +280,11 @@ Based on mlx-audio benchmarks and community reports: These are estimates. Actual benchmarks will be in `docs/overview/performance.md` after Phase 6. -## Risks & Mitigations - -### Risk: MLX stability issues -**Mitigation:** Keep PyTorch backend, allow manual override to force PyTorch on Apple Silicon. - -### Risk: Model compatibility issues -**Mitigation:** Phase 6 testing validates both backends produce similar outputs. Pin mlx-audio version. - -### Risk: mlx-audio development stalls -**Mitigation:** Project is actively maintained by Blaizzy with frequent updates. If it dies, we keep PyTorch and remove MLX in a later release. - -### Risk: Users download wrong installer -**Mitigation:** Auto-detect architecture on download page. Show big warning if mismatch detected. - -### Risk: Increased maintenance burden -**Mitigation:** Backend abstraction keeps shared logic common. Test suite validates both backends. Most work is one-time (setup), not ongoing. - ## Open Questions -- **Should we support Qwen3-ASR (MLX-only) in addition to Whisper?** Adds another model option but increases complexity. Probably phase 8+. -- **Should we backport streaming to PyTorch?** Would require chunking and callback-based generation. Probably not worth it given mlx-audio already has it. -- **What's the auto-update UX for migrating PyTorch→MLX users?** Needs design. Don't want to force reinstall, but also want to make upgrade obvious. +- **Should we support Qwen3-ASR (MLX-only) in addition to Whisper?** Adds another model option but increases complexity. Probably phase 8+. - Sure +- **Should we backport streaming to PyTorch?** Would require chunking and callback-based generation. Probably not worth it given mlx-audio already has it. - No +- **What's the auto-update UX for migrating PyTorch→MLX users?** Needs design. Don't want to force reinstall, but also want to make upgrade obvious. - it just updates, users see nothing - **Do we expose backend selection in settings or hide it?** Leaning toward auto-detect only, with env var override for power users. ## Success Metrics