Features

FlagScale provides a unified toolkit covering the complete lifecycle of large language models, multimodal models, and embodied AI models. It integrates multiple open-source backend engines under a single configuration and CLI interface, enabling consistent operation across diverse chip vendors. Key features include:

• Support for training, reinforcement learning, and inference workflows through collaboration with FlagOS plugin repositories (Megatron-LM-FL, TransformerEngine-FL, VeRL-FL, and vllm-plugin-FL)

• Scalable cross-chip communication through collaboration with FlagCX

• Embodied AI workload support through collaboration with FlagOS-Robo

• Automatic parallelism optimization that searches and selects the optimal combination of parallelism strategies, optimization methods, and memory configurations for different models, cluster setups, and chip architectures — enabling users to achieve the best model parallel training and inference performance in a fully automated, zero-effort manner