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
MoE
4 atoms · 跨 3 天 · 首见 2026-04-26 · 最近 2026-05-22
三色: 🟦 fact 0 · 🟥 take 4 · stance ▲1/▼2/◆1
来源: X 4
时态: aging:2 · fresh:2
标签: 好观点:3 · 好思考:1 · 好问题:1
叙事 (narrative) (4)
- 2026-05-22 · 探索序列级负载均衡方法
This article explores how to achieve sequence-level load balancing without incurring any loss penalty. Starting from the original Quantile Balancing (QB), we gradually derive a new method called Moving Quantile Balancing (MQB), which successfully accomplishes this objective.
tags: 好思考·好问题 · → daily
- 2026-05-22 · quantile balancing转移瓶颈到不同层
quantile balancing sounds cool until u realize ur just shifting the bottleneck to a different layer
tags: 好观点 · → daily
- 2026-05-10
[老化中] · 未来趋势判断
但我觉得模块化模型在“稀疏廉价”之外有自己的未来
tags: 好观点 · → daily
- 2026-04-26
[老化中] · 模型架构复杂性问题
Not necessarily everyone, since MoE can also add a lot of headache
tags: 好观点 · → daily
时间轴 (近 20)
- 2026-05-22 ·
narrative · 探索序列级负载均衡方法 · → daily
- 2026-05-22 ·
narrative · quantile balancing转移瓶颈到不同层 · → daily
- 2026-05-10 ·
narrative · 未来趋势判断 · → daily
- 2026-04-26 ·
narrative · 模型架构复杂性问题 · → daily
← 实体目录 · 系统日志