async 4 discloses data mix, MoE router crash sparks discussion

async 4 unusually disclosed the exact percentages of its training data mix, earning praise while showing surprising performance in a 30-step run. Discussants also dissected a training mishap where the MoE router crashed unfrozen despite R3 processing, with multimodal training suspected as the cause.

2026-09-22 ~ 2026-09-22 · 3 related posts

• Commenter praises async 4 for disclosing training data mix percentages — stochasticchasm · 2026-09-22
• Router collapse mid-run: blogger probes rare data-mix disclosure in 30-step training — stochasticchasm · 2026-09-22
• MoE training failure details: router collapsed without freezing, multimodal experts suspected — stochasticchasm · 2026-09-22