Xiaohongshu Open-Sources Training Engine, RelaX AI Circle Adds Another Major Player

On April 15, the Xiaohongshu AI platform team made a notable move in the tech community by quietly open-sourcing a reinforcement learning training engine for large models, named Relax.
Relax is built specifically for multimodal and agentic scenarios, meaning it goes beyond text to handle inputs like images, audio, video, and more — all unified and flexibly integrated. This design aligns with the prevailing AI trend, where multimodal and agent capabilities are widely seen as the next major frontier.
Technically, Relax introduces two core mechanisms: modal-aware parallelism and end-to-end asynchronous pipelining. The former intelligently allocates computing resources based on the characteristics of each modality, while the latter minimizes waiting and idle time during training through an asynchronous pipeline. Together, they boost the efficiency and scalability of multimodal training — offering practical engineering value for AI teams handling large-scale workloads.
Notably, the open-source move itself is significant. Xiaohongshu is not traditionally an AI infrastructure company. By voluntarily releasing its internal training engine, the company showcases its deep expertise in AI engineering and extends an olive branch to the developer community — leveraging technical contributions to build ecosystem influence. This is a path more tech companies are taking in the AI era.
In the AI arms race, Xiaohongshu's move came as a surprise.
Related article
Apple agrees to $250M settlement over delayed Siri AI rollout
Apple has agreed to pay $250 million to resolve a class-action lawsuit regarding its marketing of AI features prior to the iPhone 16 launch, as first reported by the Financial Times.The lawsuit claimed Apple overstated the capabilities of Apple Intel
Ando Launches Team Messaging App Bridging Human and AI Collaboration
In 2025, as Sara Du assisted companies in developing MCP servers, a recurring question emerged: how could teams leverage AI agents directly within Slack? The path was obstructed by significant technical barriers, including the complexity of message s
Sequoia-backed Empirik raises $21M to forecast outages before they strike
Before joining Sequoia Capital in 2020 as chief digital and information officer, Avon Puri spent over a decade running infrastructure at Rubrik and VMware. Three years ago, as large language models began showing their true potential, Puri, alongside
Related Special Topic Recommendations
Comments (1)
0/500

On April 15, the Xiaohongshu AI platform team made a notable move in the tech community by quietly open-sourcing a reinforcement learning training engine for large models, named Relax.
Relax is built specifically for multimodal and agentic scenarios, meaning it goes beyond text to handle inputs like images, audio, video, and more — all unified and flexibly integrated. This design aligns with the prevailing AI trend, where multimodal and agent capabilities are widely seen as the next major frontier.
Technically, Relax introduces two core mechanisms: modal-aware parallelism and end-to-end asynchronous pipelining. The former intelligently allocates computing resources based on the characteristics of each modality, while the latter minimizes waiting and idle time during training through an asynchronous pipeline. Together, they boost the efficiency and scalability of multimodal training — offering practical engineering value for AI teams handling large-scale workloads.
Notably, the open-source move itself is significant. Xiaohongshu is not traditionally an AI infrastructure company. By voluntarily releasing its internal training engine, the company showcases its deep expertise in AI engineering and extends an olive branch to the developer community — leveraging technical contributions to build ecosystem influence. This is a path more tech companies are taking in the AI era.
In the AI arms race, Xiaohongshu's move came as a surprise.
Apple agrees to $250M settlement over delayed Siri AI rollout
Apple has agreed to pay $250 million to resolve a class-action lawsuit regarding its marketing of AI features prior to the iPhone 16 launch, as first reported by the Financial Times.The lawsuit claimed Apple overstated the capabilities of Apple Intel
Ando Launches Team Messaging App Bridging Human and AI Collaboration
In 2025, as Sara Du assisted companies in developing MCP servers, a recurring question emerged: how could teams leverage AI agents directly within Slack? The path was obstructed by significant technical barriers, including the complexity of message s
Sequoia-backed Empirik raises $21M to forecast outages before they strike
Before joining Sequoia Capital in 2020 as chief digital and information officer, Avon Puri spent over a decade running infrastructure at Rubrik and VMware. Three years ago, as large language models began showing their true potential, Puri, alongside





Home






