- 顿玄2026-09-19
mimo.xiaomi.com
mimo-v2.6 RL
Training metrics of the mimo-v2.6-pro and mimo-v2.6-flash reinforcement-learning runs, live from the trainer's logs.
flash的显示stop了，这是结束了吗？
评测分数还是在25轮的，不知道更新完分数能到多少
主楼 - undefined2026-09-19
the pro run restarted at step 17 due to a GPU OOM issue caused by expert load imbalance. we have adjusted the training parallelism strategy.
咦，服务器gpu爆内存了吧。 GPU OOM
- jcc2026-09-19
- nyavana2026-09-19
炸了几次，原来连大厂也出这种问题，我放心了不是我菜 ^-^
- fqy2026-09-19
不知道欸 可能服务器炸了吧 绷不住了
- Kevin-linuxdo2026-09-19
炸显存了很正常，应该是MOE专家切的有问题导致有节点的专家大量调用把显存炸了
- mukong2026-09-19
