Ethanfel c8e6b91f67 feat: add alpha_scale_sweep to fix LoRA noise contamination ...

Previous sweep used alpha=rank (scale=1.0) which at rank 128/256 drowned
base model priors — spectral flatness went from 0.013 (baseline) to 0.094.
This sweep tests alpha dramatically below rank across r16/r32/r128 to find
the scale where LoRA nudges rather than overwrites.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

2026-04-08 17:55:05 +02:00

..

alpha_scale_sweep.json

feat: add alpha_scale_sweep to fix LoRA noise contamination

2026-04-08 17:55:05 +02:00

eval_r128_candidates.json

feat: add eval_r128_candidates.json

2026-04-08 17:28:28 +02:00

r64_overnight.json

feat: r64_overnight sweep — focused rank-64 ablation at 8000 steps

2026-04-08 01:32:23 +02:00

r128_sweet_spot.json

feat: add cosine LR decay schedule to trainer and scheduler

2026-04-08 13:25:01 +02:00

tier1_sweep.json

fix: point data_dir to features/ subdir where .npz and audio live

2026-04-08 00:45:32 +02:00

tier1_thorough.json

feat: sweep resume + 5 additional experiments (LR, target, extended)

2026-04-08 00:59:16 +02:00