Build an open-source technology
system for large-scale models
system for large-scale models
FlagOpen
Open-source Full-stack Technical Foundation for Large Models Supporting Multiple Frameworks on Heterogeneous Chips
Total Downloads of Open-source Models
640 Million Times
Total Downloads of Open-source Project Code
1400,000 Times
Language Model
Visual and Multimodal Model
Embedding Model
Embodied Model
WuDao Aquila
Tele-FLM
Emu3
EVA/EVA-CLIP
SegGPT
Omnigen
Video-XL
BGE
RoboBrain
Data
FlagData
Large model data processing toolset
Algorithm
FlagAI
Integrate multiple model algorithms
Evaluation
FlagEval
Open platform for large model evaluation
Systemware（FlagOS）
FlagScale
Efficient parallel training framework supporting heterogeneous AI chip training
FlagGems+FlagAttention
Triton-based unified operator library for large models
FlagPerf
AI chip evaluation tool
FlagCX
统一通信集合库
FlagRelease
大模型自动发版平台
Read more
FlagOS
Intelligent Computing Cluster Operating System for Large Models supporting Heterogeneous Computing Power
AI Accelerator Cards
5600+
Team Size
50+
SLA Exceeding
99.6%
AI Chips
13
Large Model Training Tasks
Large Model Evaluation Tasks
Large Model Inference Deployment Tasks
FlagOS
AI Large Models
(Language Large Models, Multimodal Large Models, Embodied Large Models, etc)
Deep Learning Framework
(PyTorch, PaddlePaddle, etc)
FlagOs: System Software Stack for Multiple AI Chips
FlagGems:
General Large Model Operator Library
FlagScale:
Training and Inference Parallel Framework
FlagTree:
Unified Compiler
FlagCX:
Unified Communication Library
Backend Compiler A
Low-level Communication Library A
Chip A
Backend Compiler B
Low-level Communication Library B
Chip B
Backend Compiler C
Low-level Communication Library C
Chip C
FlagEval
A Scientific, Authoritative, Fair and Open Large Model Evaluation System
Covering over 800+ open and closed source models worldwide
50+
Capabilities
4
Major Evaluation Areas
20+
Evaluation Tasks
Read more
FlagData
A Chinese high-quality multimodal open-source dataset with a leading global scale
Rich data resources
Wide industry coverage, diverse task types, and high data quality
Full-process data tools
Streamline the whole data processing flow from raw data to training data
Deep integration of computing power platforms
Support one-click push of data to computing power clusters to start model training
Read more
