以下部分引用 AI 总结,现阶段 AI 仍然有幻觉。请以内容的原文为准。
数据新鲜度群聊 07-14 ✓卖方 07-14 ✓Wrap 断档15天 ✕News 07-15 ✓X 断档12天 ✕页面生成 07-15
MiMo-V2-Flash
6 atoms · 跨 3 天 · 首见 2026-04-27 · 最近 2026-07-02
三色: 🟦 fact 6 · 🟥 take 0 · stance ▲0/▼0/◆4
来源: X 6
时态: fresh:6
标签: 好数字:4 · 好观点:1 · 好信源:1
事实 (fact) (6)
- 2026-07-02 · 激活参数量
MiMo-V2-Flash only activates 15B parameters at a time
tags: 好数字 · → daily
value: qty=15B
- 2026-07-02 · 预训练数据量
They pulled this off by pre training on a staggering 27 trillion tokens
tags: 好数字 · → daily
value: qty=27 trillion tokens
- 2026-07-02 · 推理加速比
unlocking a massive 2.6× speedup
tags: 好数字 · → daily
value: qty=2.6×
- 2026-05-01 · 使用MOPD进行后训练
MiMo-V2-Flash: starts from a general SFT model and uses domain-specific models from across the post-training pipeline as teachers for MOPD at the final stage of post-training, where teachers are selected heuristically by domain.
tags: 好信源 · → daily
- 2026-04-27 · original training token count
MiMo-V2-Flash (which they've rebranded to just "MiMo-V2/V2.5 now) originally had 27T
tags: 好数字 · → daily
value: qty=27T
- 2026-04-27 · rebranded name
MiMo-V2-Flash (which they've rebranded to just "MiMo-V2/V2.5 now)
tags: 好观点 · → daily
时间轴 (近 20)
- 2026-07-02 ·
fact · 激活参数量 · → daily
- 2026-07-02 ·
fact · 预训练数据量 · → daily
- 2026-07-02 ·
fact · 推理加速比 · → daily
- 2026-05-01 ·
fact · 使用MOPD进行后训练 · → daily
- 2026-04-27 ·
fact · original training token count · → daily
- 2026-04-27 ·
fact · rebranded name · → daily
← 实体目录 · 系统日志