Open-source library adds reproducible seeded optimizers, recursive init, hard negative mining
leland_mcinnes · x · 2026-09-30
lelandmcinnes details technical progress on a library's 0.6dev branch:
- New optimizers allow fixing a random seed while keeping all cores working, still yielding reproducible results
- A new initialization method called "recursive"
- Hard negative mining via negativesamplerange
Changes live on the 0.6dev branch, mostly documented in docstrings but not on by default, and may break at any time before the final release.
Related event: UMAP Gains Reproducible Optimizer and Recursive Initialization(2 posts)→
More from Research
- Math advisory group issues responsible-release guidelines for AI-generated math; Gowers amplifies, critics push back — RexDouglass · 2026-09-30
- 200 simulated AI residents stress-tested a support bot: the LLM broke 6 of 8 promises — Far-Palpitation-139 · 2026-09-30
- WhiteMatter: All-to-All Cross-Layer KV Sharing Matches Bigger Models With Half the Cache — INK-USC · 2026-09-30
- NVAlign: Direct-Gradient Post-Training Improves Non-Verbal Control in Flow-Matching TTS — BlandAIOrg · 2026-09-30
- KL Divergence Isn't Needed for On-Policy Distillation, Only Token Update Direction Matters — Wenze Lin · 2026-09-30
- Sherry-style 3:4 ternary weights hit 1.375 bits: 1.6MB WebGPU model matches 7.8MB int8 — Brilliant-Hall1387 · 2026-09-30