Yandex/AliceAI 80B-A3B open post-training update: loss curves and a live training stream
jjusko20 · reddit · 2026-10-01
A Reddit user shared progress on post-training the Yandex/AliceAI Foundation 80B-A3B model, posting loss curves (sampled from the last micro of each step to explain variation) and noting help from Gemini 3.8 Flash High. The training is being live-streamed via a Cloudflare tunnel, continuing their "watch me post-train" thread.
More from Research
- SimWorld Benchmark: 35.8% of 3D Scene Edits Break Something Else — ZhitingHu · 2026-10-01
- Evaluating One Model on ProgramBench Now Costs Over $10k — jyangballin · 2026-10-01
- One-click Fukui Index calculator: SMILES in, reaction-site heatmaps out — CatAstro_Piyush · 2026-10-01
- How GPT-6 Astra solved ARC-AGI-3: an interactive move-by-move breakdown — GregKamradt · 2026-10-01
- New study: more compute doesn't mean better accuracy across agent harnesses — xiye_nlp · 2026-10-01
- LongHarness benchmark: same LM, >10x efficiency gaps across long-context agent harnesses — xiye_nlp · 2026-10-01