
9 月 10 日上午唐杰在 X 引用转帖,公开质疑简单根据能力倒推 Fable 5 规模的假设,指出最优模型规模取决于数据量、激活参数量、强化学习环境数与目标推理成本,并发布了教师节寄语与微博动态。
“Are you sure? Finding the optimal model size is tricky: data volume, active parameter count, the number of environments, and the target inference cost. Model performance also depends on many other factors, each introducing its own variability.” 1
| 评估维度 | Charlie O'Neill 的外部推断假设 | 唐杰指出的工程约束核心变量 |
|---|---|---|
| 参数口径 | 关注总参数量(推测 Fable 约为 2-2.5T,对比 K3 的 2.8T) | 强调激活参数量(Active Parameters),在稀疏 MoE 架构中起决定作用 |
| 数据与环境 | 宽泛提及拥有更多算力与更好优化器 | 明确区分数据总量(Data Volume)与强化学习环境数量(Environments) |
| 落地约束 | 主要以基准能力和同级别表现作为倒推依据 | 将目标推理成本(Target Inference Cost)作为定义最优规模的硬性前提 |
| 确定性认知 | 认为能力相近且算力更充裕时参数必然更小 | 强调影响性能的变量极多,每一项都在引入独立波动 |
x.com
x.com
x.com
weibo.com
weibo.com
x.com
x.com
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.