Tinker ships Dimuon, a LoRA optimizer that beats AdamW in fewer steps on SFT tests
miramurati · x · 2026-10-15
Mira Murati amplified two Tinker updates: the launch of Dimuon, a new optimizer built for LoRA fine-tuning that often reaches lower loss than AdamW in fewer steps in their SFT tests (still experimental, usable today in Tinker with an accompanying optimizers guide), and an ongoing hiring push for researchers and infra/systems experts.
Related event: Thinking Machines Releases Dimuon Optimizer for LoRA Fine-Tuning(4 posts)→
More from Companies & People
- NVIDIA to showcase neural rendering, physical AI and robotics at SIGGRAPH Asia 2026 — tomaszbednarz · 2026-10-15
- Anthropic's Leaked IPO Papers Show $518B in Spending Commitments, 80% Non-Cancelable — examachine · 2026-10-15
- Meta's superintelligence lab runs 200+ direct reports under Alexandr Wang, by design — CurieuxExplorer · 2026-10-15
- Tolan quietly wins millennial women; marketers say tech should stop chasing 22-year-olds — ajaymehta · 2026-10-15
- rakyll: Google's founders built a self-correcting culture unlike any other big tech firm — rakyll · 2026-10-15
- Webinar: Brand Agents and How AI Is Becoming the New Shopping Front Door — annetgriffin · 2026-10-15