Gradient 华语社区

构建开放智能基础设施。Lattica | Parallax | Echo
关注 @Gradient_HQ 获取更多资讯。

[Gradient @Gradient_HQ]

Proud to share MERA: joint work from a group of brilliant researchers including our very own.

It evolves small models with SkillBook + routing + adapters so they take on more of the agent workload over time.

Paper + code 👇

[Gradient @Gradient_HQ]

Agents are impressive. Multi-agent systems are where things get interesting.

As AI systems evolve from single agents into networks of specialists, a new question follows:

How do agents decide who does what?

That’s a problem we’ve been exploring with Symphony.

Symphony brought multi-agent collaboration across heterogeneous environments. To take that further, we released Symphony-Coord, where coordination and specialization adapt through task context, performance, and feedback.

https://t.co/7qdcmO40RU

[Bill @Bill58861938368]

The Bitter Lesson of Asynchronous RLHF:
Why Staleness Matters and How to Control It https://t.co/DSh91oNn4Z

[Gradient @Gradient_HQ]

A self-evolving agent + a 428B model + 3 Macs = ?

Your own AI lab.

We ran @MiniMax_AI M3 locally with @tryParallax, right on our desk.

Then @GA_agent_ai took over to create a 5-stock portfolio and write it to disk.

No cloud. No API bills. Nothing left the machine.

Wild to see a ~3K-line agent drive all this with a 400B+ model on local hardware.

Thanks to the GenericAgent and MiniMax teams for making local AI feel real.

[MiniMax (official) @MiniMax_AI]

This is a glimpse of where local AI is heading and we are glad to be part of it.

Really impressive work by all the teams involved @Gradient_HQ, @tryParallax, and @GA_agent_ai

[Parallax @tryParallax]

nvidia going all in on local ai.

here's our take: it shouldn't depend on which chip you bought.

sparks, macs, the 5090 already on your desk, we cluster across all of it and split your favorite model pipeline-parallel so it runs fully private and local.

[Gradient 华语社区 @Gradient_CN]

我的怎么样? 👀

[Eric @0xEricYang]

We're hiring at Gradient.

Building open-source environment infrastructure for our distributed RL training stack — reproducible, scalable to thousand-GPU runs

Looking for 1–2 RL Environments engineers / tech leads: You've designed verifiers, built sandboxes for agentic RL rollouts, or shipped RL training data pipelines that survived contact with real training.

Domain depth in math, code, agent, tool, or GUI is a plus. PhD not required.

Also hiring research interns: PhD / Masters students with hands-on RLHF / RLVR / GRPO / DPO / agentic RL experience. Open-source footprint matters more than paper count. Most intern roles convert post-grad. No age cap. Founding-team-level equity for the right people.

DMs open.

[Gradient 华语社区 @Gradient_CN]

@Alibaba_Qwen 🤍

[Gradient 华语社区 @Gradient_CN]

非常高兴看到 Parallax 赋能 Theta 的愿景!

我们的框架能够让分布式机器协同工作，在全球范围内以高性价比扩展 AI 推理能力。

Parallax 已经为智能 Agent 时代做好准备了！

[Theta Network @Theta_Network]

To make this work, we adapted Parallax, @Gradient_HQ's distributed inference framework, to run across EdgeCloud's global node network. One API endpoint, model split across many machines, no centralized cluster required.

[Viviennn ★ @whereisvivien]

「アグモン、今だ！超進化！」 👾👾👾

[Gradient 华语社区 @Gradient_CN]

@CamilaQuirinoMH 👀

[Viviennn ★ @whereisvivien]

把这个海报发给朋友的时候，他们都会问为什么会有「数码宝贝」

可能只是小时候日本动漫看多了，看到什么 evolving、进化这种词，脑袋一拍就觉得主题怎么都得定数码宝贝

虽然海报卡通 但是活动内容是包硬核的

我们邀请到了
- OpenClaw-RL 的作者
- 上海人工智能实验室研究员
- MiniMax 技术专家
- Unipat AI 的 Cofounder
- 红杉中国 & 真格基金投资人
- Gradient 的 AI Engineer 等等嘉宾

从模型训练、RL infra、online learning 等不同角度来一起探讨 Self-evolving

这次活动上基本不会有任何产品硬广 主要就是给各路 RL 大神们搭个台子 让大家尽情表演。也非常感谢各位合作方爸爸们的支持！

以及非常欢迎这周在上海参加小红书黑客松的各路大神们来玩 如果有想要在现场 demo 的 我举双手赞成！！！

Luma链接：https://t.co/2BVmqbifJy

[Gradient 华语社区 @Gradient_CN]

@lingrooo

[Gradient 华语社区 @Gradient_CN]

那些考察模型已记忆内容的基准benchmark正在快速饱和。ARC-AGI-3提出了一个更难的问题：AI是否真的能够在现场实时学习��事物？

我们一直在探索的一个方向是：multi agent orchestration。在我们的研究中，通过多个回合协调四个前沿大语言模型，其表现能够稳定地匹配或超越最强的单一模型，甚至在它们任何一个单独都无法解决的任务上也是如此。

“最佳单一模型”与“模型最佳协同”之间的差距，正是隐藏着大量真正进步的地方。

关于我们这项multi-turn、multi-agent orchestration研究的更多资料：https://t.co/uSIy9z70Z1

[ARC Prize @arcprize]

Announcing ARC-AGI-3

The only unsaturated agentic intelligence benchmark in the world

Humans score 100%, AI <1%

This human-AI gap demonstrates we do not yet have AGI

Most benchmarks test what models already know, ARC-AGI-3 tests how they learn

[Gradient 华语社区 @Gradient_CN]

很高兴看到multi-agent system终于开始受到严肃的工程级重视了。

我们一直很关注的一点是：随着智能体能力越来越强，编排层（orchestration layer） 的重要性已经和模型本身不相上��。

我们的 Symphony 项目就是在探索：如果完全去掉中央控制器，让智能体在消费级硬件上通过去中心化任务分配 + 加权投票的方式自主协作，会发生什么。

结果显示，我们在集中式框架（如 AutoGen 等）的基础上实现了最高 41.6% 的准确率提升，同时在普通消费级 GPU 上运行，编排开销不到 5%。

想了解更多细节，欢迎阅读我们的 Symphony arXiv： https://t.co/WPPfntapKQ

[Anthropic @AnthropicAI]

New on the Anthropic Engineering Blog:

How we use a multi-agent harness to push Claude further in frontend design and long-running autonomous software engineering.

Read more: https://t.co/HWvmXk1ykn

[Viviennn ★ @whereisvivien]

. @FireworksAI_HQ 最近发了一篇文章
Frontier RL Is Cheaper Than You Think —— 强化学习的拼多多时刻
https://t.co/zUXYvamedy

这篇文章其实就是在回溯 他们帮助Cursor 训练了基于 Kimi K2.5 的新模型 Composer 2的技术框架。

他们发现RL训练不需要GPU集群，用异步架构和压缩传输就能把分布式 rollout 跑起来，从而大幅降低成本。

二月份的时候我刚好写了一篇文章，解释了一下@Gradient_HQ 的Echo-2 为什么选择去中心化框架来做RL 核心想法其实和Fireworks是完全一样的：

把 RL 训练中最耗资源的 Rollout 阶段，从中心化集群里解耦出来，分发到异构设备上。

RL 本质上是两个阶段在交替进行：一个是 Rollout——模型在环境中采样、做推理、跟环境��互；另一个是 Optimization——拿到采样数据之后做反向传播更新参数。这两个阶段对算力的需求完全不同，节奏也不同。这会直接导致 GPU 有大量时间处于等待，算力空转。

而在 RL 训练中，80% 以上的算力消耗其实集中在 Rollout 阶段。这个阶段天然具备高并行属性，对网络延迟的容忍度很高。不需要在同一个机房，不需要统一的芯片规格，任何能跑推理的设备都可以参与。

所以说，难道分布式训练的春天真的要来了？🤔

[Gradient 华语社区 @Gradient_CN]

@Linazhaozhao2 爱parallax

[Gradient @Gradient_HQ]

Dobby is a free elf now.

Open models, open orchestration, open compute. The agentic RL stack that used to live inside walled gardens just showed up on hardware you can order and frameworks you can fork.

No masters needed.

[Gradient @Gradient_HQ]

Our GTC takeaway is clear: NVIDIA is betting hard on open.

- NemoClaw turns OpenClaw into enterprise infrastructure.
- Nemotron 4 will be open-sourced.
- Nemotron Coalition puts eight labs on a shared open frontier model.

This is what we've been building toward. Open infrastructure for open intelligence is the direction the biggest AI companies are taking.