Beam LoRA fine-tuning benchmark: 500 steps, 16M tokens, 16 minutes on one 8x B300 node
QGallouedec · x · 2026-10-16
Follow-up benchmarks for LoRA SFT of Reflection AI's Beam (501B/23B active) on a single 8x B300 node: 500 steps over a reasoning subset of Mixture-of-Thoughts (16M tokens) take just 16 minutes, with training loss dropping from 0.89 to 0.70 within the first 30 steps and token accuracy rising from 73% to 77%. One epoch of Multilingual-Thinking trains in under 2 minutes.
More from Research
- Hugging Face launches AI4Science heatmap tracking open models from 79 orgs, AI2 and OpenMed lead — anshulkundaje · 2026-10-16
- MIT's Neil Thompson: AI comes for tasks before jobs, and fixed costs favor big firms — bigdata · 2026-10-16
- UMD's PAIR framework bridges perception and action in VLA models via bridge tokens — Kaixi Feng · 2026-10-16
- CI-MSE offline metric hits -0.87 Spearman correlation with real robot task success vs weak raw MSE — chris_j_paxton · 2026-10-16
- Colossal Biosciences uses AI to decode genetic switches behind the extinct dodo — castrotech · 2026-10-16
- Colossal Biosciences uses AI to identify gene switches in dodo de-extinction push — castrotech · 2026-10-16