Oneflow
OneFlow builds an open-source deep learning framework and supporting cloud, MLOps, and training platforms for AI developers, researchers, and enterprise AI teams in China, featuring a Global View/SBP distributed training architecture designed for large neural networks.
- Company typePrivate
- Founded2017
- HeadquartersHaidian, China
- Headcount101–250
- GTM typeB2B
- OfferingSoftware
What Oneflow does
OneFlow (OneFlow Yiliu Technology Co., Ltd. / 一流科技有限公司), founded in January 2017 in Haidian District, Beijing, is a private Chinese deep learning infrastructure company with 101-250 employees. Its core product is the OneFlow open-source deep learning framework, distributed via GitHub and PyPI under a PyTorch-compatible API, targeting AI developers, researchers, and enterprise AI teams that need to train large neural networks across distributed GPU clusters. The framework's defining architectural contribution is a Global View programming paradigm built around a Global Tensor abstraction paired with SBP (Split, Broadcast, Partial) signatures and automatic boxing, which the company claims allows users to program distributed systems as if they were a single supercomputer. Four named proprietary technologies underpin the system: automatic data-model hybrid parallelism, static scheduling, decentralization, and full-link asynchronous streaming execution, all oriented toward solving memory-wall constraints at the cluster level.
The product portfolio has expanded beyond the framework to three commercial offerings launched in 2021: OF Smart Cloud (a cloud AI development platform supporting public, private, and hybrid deployment), OneBrain (an enterprise MLOps platform covering data preparation, model building with lineage tracking, deployment, DevOps, and monitoring on a CI/CD/CD pipeline), and OneLab (an AI training community platform with online programming environments and GPU computing). Supporting ecosystem libraries include FlowVision (a torchvision-equivalent computer vision library) and libai (large-model pretraining). The framework integrates with NVIDIA Triton Inference Server for serving and provides an ONNX export path for cross-platform inference.
The business model combines a free open-source framework as a developer acquisition channel with subscription-style monetization on the cloud and enterprise platforms. Public pricing on cloud and enterprise tiers is not disclosed, and no enterprise customer logos or revenue figures have been publicly disclosed. Total disclosed venture funding is approximately $4.92M across three rounds from May 2017 to February 2019, with Kuaishou Technology as the most strategic disclosed investor. Distribution runs through PyPI, GitHub, Bilibili video content, a QQ developer community, technical documentation, and direct enterprise sales in China.
Oneflow firmographics
Firmographics- Name
- Oneflow
- Legal name
- OneFlow一流科技有限公司
- Website
- https://oneflow.org
- Company type
- Private
- Founded year
- 2017
- Operating status
- Operating
- Headcount range
- 101–250 employees
- Short description
- OneFlow builds an open-source deep learning framework and supporting cloud, MLOps, and training platforms for AI developers, researchers, and enterprise AI teams in China, featuring a Global View/SBP distributed training architecture designed for large neural networks.
- Ownership category
- akta.pro rank
Oneflow industry classification
Industry- Product category
- Deep Learning Framework / AI Infrastructure Software
- NAICS
- Computer Systems Design and Related Services (54151), Software Publishers (5132), Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (518)
- SIC
- Services-Prepackaged Software (7372), Services-Computer Programming, Data Processing, Etc. (7370)
- akta.pro primary industry
- End-to-End MLOps & ML Platform Suites (HDAAABAA)
- akta.pro secondary industries
- Model Deployment, Serving & Inference Platforms (HDAAABAF), Pipeline Orchestration & Workflow Automation (HDAAABAD), Enterprise Foundation Model Integration & APIs (Connectors, Governance, Deployment) (HDAAACAO)
Keywords
Where Oneflow is headquartered
LocationHeadquarters
- HQ city
- Haidian
- HQ country
- China
- HQ region
- Asia
Offices2 records
Markets served
Oneflow business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Personnel, Technology or R&D, Infrastructure, Marketing or Sales, Operations
Revenue model
- OF智能云 (OF Smart Cloud): Cloud-based AI development and training platform providing services including AI training, AI development, GPU computing power, and model deployment. Supports public cloud, private cloud, and hybrid cloud deployment modes.
- OneBrain MLOps Platform: Enterprise-grade MLOps platform offering end-to-end AI lifecycle management from data preparation to model deployment and monitoring.
- OneLab AI Training Platform: AI training and practice community for developers providing online programming environments, GPU computing, and open-source algorithm tutorials.
- Open Source Framework (Community): The OneFlow deep learning framework is open source and freely available, generating community adoption that may lead to enterprise licensing or cloud service adoption.
Go-to-market motion2 records
Distribution channels6 records
Marketing channels6 records
Oneflow product offering
Product offeringCore offering
OneFlow develops and distributes an open-source industrial-grade general-purpose deep learning framework featuring a Global View programming paradigm and SBP-based distributed training, with near-zero-cost PyTorch migration. It complements the framework with OneBrain (enterprise MLOps platform for end-to-end AI lifecycle management), OneLab (AI training and practice community with GPU compute), and OF Smart Cloud (a cloud AI training/development platform supporting public, private, and hybrid deployment).
Product overview
OneFlow一流科技 offers a suite of AI infrastructure products centered on the OneFlow深度学习框架 (OneFlow Deep Learning Framework) as the core engine. The product portfolio consists of OneFlow (the open-source general-purpose deep learning framework), OneBrain (a one-stop MLOps platform for enterprise AI development lifecycle management), OneLab (an AI training and practice community), and OF Smart Cloud (a cloud platform supporting public, private, and hybrid deployment). The framework supports distributed training, model serving via Triton, ONNX export, and PyTorch model compatibility.
Differentiator
Problem solved
Functional benefit
Products and services
- OneFlow Deep Learning Framework (OneFlow深度学习框架) An open-source, industrial-grade general-purpose deep learning framework providing both dynamic (Eager) and static (nn.Graph) execution with a PyTorch-compatible API. Core innovations include Global Tensor, SBP signatures, automatic boxing, data/model/pipeline/hybrid parallelism, AMP, ZeRO, activation checkpointing, and OneEmbedding for large-scale recommendation systems. Targeted at AI developers, researchers, and enterprise AI teams building distributed deep learning workloads.
- OneBrain Data Intelligence Platform (OneBrain数据智能平台) A one-stop MLOps platform providing end-to-end AI lifecycle management for data scientist teams: data preparation, model building with lineage tracking, model deployment, DevOps, and model monitoring. Supports elastic scaling, reinforcement learning, and large-scale advertising recommendation systems. Targeted at enterprise AI teams standardizing their AI development and deployment pipelines.
- OneLab AI Training Platform (AI实训平台) An AI training and practice community for AI developers offering online programming environments, GPU computing resources, open-source algorithm explanations, and quick model creation/deployment capabilities. Aimed at becoming a leading community and exchange platform in the large-model domain.
- OF Smart Cloud (OF智能云) A cloud-based AI development platform offering services ranging from beginner AI talent development to enterprise MLOps integration management. Includes AI training (OneLab) and AI development (OneBrain) offerings, supporting public cloud, private cloud, and hybrid cloud deployment models, with curated algorithm tutorials, open-source algorithm testing, and online programming capabilities.
Quantifiable outcome
- Training speed faster than NVIDIA HugeCTR for Wide & Deep recommendation models
- +3 more outcomes
Companies that use Oneflow
Customer profileSegments6 records
Ideal customer profiles4 records
Oneflow technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration4 records
AI capability9 records
Feature15 records
Oneflow partnerships and signals
Strategic signalScale indicators3 records
Recent moves6 records
Expansion highlights6 records
Oneflow competitors and assessment
Company assessmentDirect peers
- DeepSpeed (Microsoft): Microsoft's DeepSpeed library focuses on large-model training optimizations (ZeRO, ZeRO-Infinity, 3D parallelism) that overlap heavily with OneFlow's distributed primitives. Both are aimed at enabling trillion-parameter-scale model training.
- PyTorch (Meta): Meta's open-source deep learning framework is OneFlow's primary direct competitor and the ecosystem whose API OneFlow explicitly emulates. Both target distributed training of large models with comparable feature sets (DDP, AMP, ZeRO-style optimizations) and overlap heavily on developer mindshare.
- NVIDIA Megatron-LM: NVIDIA's reference implementation for training large transformer models. It overlaps with OneFlow on tensor/pipeline/model parallelism primitives and represents the performance benchmark against which OneFlow claims its own distributed superiority.
- Huawei MindSpore: Huawei's deep learning framework is the closest direct domestic Chinese competitor to OneFlow, with similar distributed-training ambitions and integration with Huawei Ascend hardware. Competes for the same Chinese enterprise and research workloads.
- TensorFlow / JAX (Google): Google's flagship deep learning frameworks compete head-to-head with OneFlow across research and enterprise. JAX in particular targets high-performance distributed training for large models — directly addressing the use case OneFlow is positioned for.
Emerging players
- MLflow (Databricks): Open-source MLOps platform for managing the ML lifecycle. Competes with OneFlow's OneBrain for enterprise model lifecycle, deployment, and monitoring workloads, especially within data-platform-centric enterprises.
- Colossal-AI: An emerging distributed training framework focused on large AI models with strong PyTorch compatibility. Overlaps directly with OneFlow's core use case of large-model training and is gaining traction in both research and industry.
- Weights & Biases: AI developer platform offering experiment tracking, model management, and MLOps tooling that overlaps with OneFlow's OneBrain platform for enterprise AI team workflows and reproducibility.
Broad incumbents
- Determined AI (HPE): Enterprise-grade deep learning training platform providing MLOps and distributed training capabilities (now part of HPE). Direct competitor to OneFlow's OneBrain MLOps platform for enterprise AI lifecycle management.
- Baidu PaddlePaddle: Baidu's open-source deep learning platform is a broad incumbent in China, offering framework plus ecosystem (PaddleNLP, PaddleCV, PaddleClas, etc.) that competes with OneFlow's full-stack (framework + MLOps + training) offering.
Market position
Strengths4 records
Weaknesses5 records
Competitive moat5 records
Key risks6 records
Key highlights6 records
Customer concentration
Oneflow social profiles
Digital presenceOneflow financial estimates
Financial estimateRevenue estimate
Valuation estimate
Oneflow leadership team
Management profileNumber of profiles
Profiles2 records
Oneflow funding detail
Funding detailFunding overview
Funding rounds3 records
Investors3 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
Oneflow M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about Oneflow
What does Oneflow do?
OneFlow develops and distributes an open-source industrial-grade general-purpose deep learning framework featuring a Global View programming paradigm and SBP-based distributed training, with near-zero-cost PyTorch migration. It complements the framework with OneBrain (enterprise MLOps platform for end-to-end AI lifecycle management), OneLab (AI training and practice community with GPU compute), and OF Smart Cloud (a cloud AI training/development platform supporting public, private, and hybrid deployment).
Is Oneflow a public or private company?
Oneflow is a private company. It is classified as venture growth investor backed and is currently operating.
When was Oneflow founded?
Oneflow was founded in 2017. It employs 101 to 250 people.
Where is Oneflow based?
Oneflow is headquartered in Haidian, China, in the Asia region.
How does Oneflow make money?
Four revenue lines are on record. OF智能云 (OF Smart Cloud) is the primary driver. The others are oneBrain MLOps Platform, oneLab AI Training Platform and open Source Framework (Community).
Who are Oneflow's main competitors?
Direct peers on record are DeepSpeed (Microsoft), PyTorch (Meta), NVIDIA Megatron-LM, Huawei MindSpore and TensorFlow / JAX (Google). Emerging players are MLflow (Databricks), Colossal-AI and Weights & Biases. Broad incumbents are Determined AI (HPE) and Baidu PaddlePaddle.
Does Oneflow have an API?
Yes. OneFlow provides the oneflow-onnx conversion tool (pip install oneflow-onnx) which enables exporting OneFlow static graph (nn.Graph) models to ONNX format. The export_onnx_model() function supports external_data for large models, opset versioning, dynamic batch size, and flow_weight_dir for weights. A companion convert_to_onnx_and_check() function validates converted models against the original OneFlow model outputs. Developer documentation is at github.com/Oneflow-Inc/oneflow_convert.
What industry is Oneflow in?
Oneflow's product category is Deep Learning Framework / AI Infrastructure Software. Its primary akta.pro industry code is HDAAABAA, End-to-End MLOps & ML Platform Suites, with a secondary code of HDAAABAF, Model Deployment, Serving & Inference Platforms. Its NAICS code is 54151 and its SIC code is 7372.