跳到正文
Hacker News (official RSS)(ricursive.com)·· 18 天前品質評分57

Ripple-65B Preview 宣稱以較少 Rollouts 達成 Agentic Coding SOTA

Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding

重點導讀

Ripple-65B Preview 在 TerminalBench 2.1 得分 80.5,作者稱其使用約為先前 SOTA Reflection Beam 1/70 的 rollouts,達到 Agentic Coding 領域的 SOTA。

AI 伴隨問答

自備 Key 深度分析

針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:

正文

Rather than investing in a brute-force RL climb, we carefully iterated on our SFT and RL recipes with a focus on compute efficiency. We opted not to buy any posttraining data, instead devising our own synthetic data pipelines, quality filtering, and coverage metrics. 

Our mixed RL recipe uses

truncated importance sampling

,

score centering

, and

linear length penalty

with high-quality long-horizon synthetic environments. The RL tasks are tuned for our model capabilities, and we continuously added harder tasks as the model improved.

The end result was a steep RL climb from a strong initial SFT checkpoint. We reached a TerminalBench 2.1 score of 80.5 with only 1.4M rollouts, roughly 70x fewer rollouts than Reflection Beam, the prior SOTA.

來源: Hacker News (official RSS) (ricursive.com)
開啟原文

讀者討論

理性交流 · 即時討論