Large-scale Agent Training Environments
Build ROCK, a large-scale agent training environment framework for post-training, evaluation, and data production, covering reinforcement learning environment development, deployment, and management.
The Intelligent Engine department belongs to Alibaba Holding Group's Platform Technology organization. It builds internal search, recommendation, advertising, and AI engineering systems. The rapid growth of AIGC brings new challenges: improving large-model training and inference performance and cost, reshaping model development workflows, and addressing new demands from online model services.
We provide a complete engineering system for group users, spanning data, training, evaluation, prediction, application development, and solutions. Working closely with the Happy family model teams, we co-develop multimodal generative large models and take on core responsibilities for Alibaba Group multimodal generation model development.
We value a strong technical culture and open sharing. The team has contributed open-source projects including ROLL, Megatron-LLaMA, XDL, and Euler, and continues to share outcomes with the broader community.
Who We Are
The Search & Governance Platform builds foundational engineering platforms for Alibaba Group LLMs and multimodal models. The team provides large-scale agent training environments for LLMs and high-performance inference services for the HappyHorse and HappyOyster model families.
Build ROCK, a large-scale agent training environment framework for post-training, evaluation, and data production, covering reinforcement learning environment development, deployment, and management.
Build foundational engineering platforms for Alibaba Group LLMs and multimodal models, supporting training, evaluation, prediction, and application development for the HappyHorse and HappyOyster model families.
Optimize the full stack for VL, Omni, DiT, Diffusion Model, and AR architectures, from low-level kernels to inference service scheduling, balancing model quality, performance, and cost.
Deliver production inference for leading multimodal understanding and generation models in AI applications, spanning inference architecture, end-to-end service orchestration, and performance optimization.
Selected Results
As the Alibaba HappyHorse inference engineering team, we bring leading multimodal understanding and generation models into AI applications, covering inference architecture design, end-to-end service orchestration, and deep performance optimization.
Open Source
An open-source LLM training framework based on Megatron, supporting efficient distributed training.
Alibaba's open-source sparse model training framework for large-scale recommendation and advertising.
An open-source RL training framework for efficient distributed RL post-training at scale.
An open-source reinforcement learning environment development framework that simplifies environment development, deployment, and management.
Alibaba's open-source distributed graph learning engine for large-scale GNN training.
A large model training framework for recommendation and advertising at industrial scale.
Open Roles
Select a role or apply button to open the Alibaba Talent page.
Application