跳到正文
Lisan al Gaib· @scaling01 · X·· 19 小时前AI 评分58
AI 导读

Thinking Machines 公开可变循环模型 TML-d24-varloop-research,用于研究循环次数对模型表现的影响。其模型卡显示,GraphWalks BFS、MRCR v2 等长上下文基准随循环次数增加而提升,AA-Omniscience、SimpleQA Verified、SimpleVQA、WorldVQA 等知识密集型基准则没有改善。作者转发时补充称循环能提升长上下文表现并不意外。

正文 · 原文

also to no surprise:
- looping improves long-context

【引用 @ziqiao_ma】Besides Inkling 2 Mini (aka Loopling), we also share a variable-looped model, TML-d24-varloop-research, because Inkling 2 Mini only supports a fixed number of looping passes.

https://thinkingmachines.ai/model-card/tml-d24-varloop-research/

Studying the variable-loop model, we saw that performances on long-context benchmarks improve with more passes, such as GraphWalks BFS and MRCR v2; but knowledge-intensive benchmarks doesn’t improve with more passes, such as AA-Omniscience, SimpleQA Verified, SimpleVQA, and WorldVQA.

We shipped Inkling 1.0 on July 15, 2026. Now Inkling 2 Mini outperforms it in almost every way at just 2.7% of the size. That’s how we iterate at Thinky :bufo-ship:

Enjoy Inkling in your pocket!

引用Martin Ziqiao Ma@ziqiao_ma
Besides Inkling 2 Mini (aka Loopling), we also share a variable-looped model, TML-d24-varloop-research, because Inkling 2 Mini only supports a fixed number of looping passes. https://thinkingmachines.ai/model-card/tml-d24-varloop-research/ Studying the variable-loop model, we saw that performances on long-context benchmarks improve with more passes, such as GraphWalks BFS and MRCR v2; but knowledge-intensive benchmarks doesn’t improve with more passes, such as AA-Omniscience, SimpleQA Verified, SimpleVQA, and WorldVQA. We shipped Inkling 1.0 on July 15, 2026. Now Inkling 2 Mini outperforms it in almost every way at just 2.7% of the size. That’s how we iterate at Thinky :bufo-ship: Enjoy Inkling in your pocket!
在 X 查看被引用的帖子

来源:Lisan al Gaib · x.com