NVIDIA发布Agora:基于Git共享内存的研究型Agent协作框架
Banger paper from NVIDIA on shared memory for research agents.
首次展示无监督多Agent自主科研协作的具体工程实现与量化成果,Agora的Git共享内存机制为集体自研究提供了可复用的新范式,值得收藏参考。
Banger paper from NVIDIA on shared memory for research agents.
NVIDIA 关于研究智能体共享内存的出色论文。
(bookmark it)
(收藏它)
If you run several coding agents on the same research problem, this design keeps them from repeating each other's experiments and lets each agent build on results the others have already verified.
如果你在同一研究问题上运行多个编码智能体,这种设计可以防止它们重复彼此的实验,并让每个智能体基于其他智能体已验证的结果进行构建。
Agora records every result, hypothesis and verification as an immutable Git commit.
Agora 将每个结果、假设和验证记录为不可变的 Git 提交。
Parent edges show what each claim builds on, and an index lists open branches and which claims have been verified.
父级边显示每个声明所依赖的基础,索引列出了开放的分支以及哪些声明已被验证。
They ran 13 LLM workers for nearly 12 days without assigned tasks or a central planner. The workers had to initialize a 119.6M-parameter hybrid model from 141 donor models without training data or gradient updates.
他们在没有分配任务或中央规划器的情况下,运行了 13 个 LLM 工作者近 12 天。这些工作者必须在没有训练数据或梯度更新的情况下,从 141 个捐赠模型中初始化一个拥有 1.196 亿参数的混合模型。
The workers posted 1,703 contributions.
工作者发布了 1,703 次贡献。
They cut the evaluator from 3.39 to 1.899 bits per byte, closing 62% of the gap to a trained GPT-2 124M. All 165 independent reproductions succeeded.
他们将评估器从每字节 3.39 位降低到 1.899 位,缩小了与经过训练的 GPT-2 124M 之间差距的 62%。所有 165 次独立复现均成功。
Paper: https://arxiv.org/abs/2609.18094
论文:https://arxiv.org/abs/2609.18094
Chat with Paper: https://academy.dair.ai/papers/agora-git-as-shared-memory-for-collective-autoresearch-2609.18094
与论文对话:https://academy.dair.ai/papers/agora-git-as-shared-memory-for-collective-autoresearch-2609.18094
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力