推文

@harden_369741 · 2026-10-12 23:53

Everyone talks about bigger models, nobody talks about the GPUs sitting idle while rollouts finish. I have watched that waste on my own runs and it hurts. QwenGyre making rollout and training share one elastic pool is the least glamorous and most important kind of progress.

曝光 20 · 评论 0 · 点赞 0 · 书签 0