ONEFLOW V0.8.0 officially released

It's today  OneFlow  Open source  717  God ,OneFlow v0.8.0  Official release . This update contains 523 individual commit, Please check the link for the complete update list ：

https://github.com/Oneflow-Inc/oneflow/releases/tag/v0.8.0

, Welcome to download and experience the new version , Looking forward to your feedback .

 

OneFlow v0.8.0  It mainly includes the following new highlights, functions and optimization ：

 

1. Compatibility

 

PyTorch compatible API  More perfect .

Add a series of and  PyTorch 1.10.0  Compatible functions and interfaces , Including new and  PyTorch  The alignment of the  68  individual  API,  Repair the  84  Operator and interface compatibility  bug, Support users will be more  PyTorch  The model is migrated to  OneFlow.

 

2.Global  Operators support

 

All operator pairs  Global Tensor  The support of is more complete and efficient .

Repair the  28  individual  Global Tensor  dependent  bug, newly added  180  individual  Global  Operator single test , Users can use  Global Tensor  Develop distributed models .

 

3. performance

 

Graph  The performance of advanced features is more mature ：

 

• In addition to the original  ZeRO-DP  outside ,ZeRO  The zero redundancy optimizer can work with  MP,2-D,3-D  Use in parallel , Further save the cost of video memory .

• Graph  Proposed a new pipelined parallel  API, While simplifying pipelined parallel configuration, accelerate pipelined parallel and  3-D  Parallel performance .

• In order to further improve  Graph.debug  Debugging efficiency , Add about logic diagram 、light plan  Physical diagram 、 Memory analysis 、Python  Multi dimensional debugging functions such as stack information .

 

 

 

be based on  OneFlow v0.8.0  and  

LiBai v0.2.0

 , GPT  and  BERT 3-D  The speed optimization under parallel is obvious

, The speed performance in multiple dimensions exceeds that in the same configuration  Megatron-LM（

See details for data ：https://libai.readthedocs.io/en/latest/tutorials/get_started/Benchmark.html

）.

 

4.OneEmbedding Components

 

OneEmbedding  It is an expansion component specially designed for large-scale recommendation system , High performance 、 Scalable 、 High flexibility .

It has the following characteristics ：

 

• Support for tiered storage , Dynamic capacity expansion  Embedding, Users can expand at a lower cost  Embedding  Capacity

• Hybrid parallel strategy , It can easily expand the model horizontally to the scene of multiple machines and multiple cards

• Communication quantization compression function , In the parallel scenario , Quantize and compress the communication data , To reduce traffic , Improve your training speed

• Efficient data pipeline , Execute the parts of the model without data dependency in advance , Overlap in time

• Support automatic hybrid accuracy training , Part of the calculation will be converted into  FP16  Data type calculation , Improve training speed while reducing the occupation of video memory , And it can ensure the convergence accuracy of the model

• Provide a series of high-performance for common operations of recommended system models  CUDA  operator

• Support flexible model building

 

API  file ：

https://docs.oneflow.org/master/cookies/one_embedding.html

 

5. Multi device adaptation

 

OneFlow  Provide a simple, efficient and extensible hardware abstraction layer  EP（Execution Provider）, To cope with the complexity of adapting to different hardware .

After introducing the hardware abstraction layer , Users do not need to pay attention to the specific implementation details of the underlying hardware and framework , Each module of the framework can be adapted to new hardware devices without modification , meanwhile , Users only need to follow the Convention of hardware abstract interface and the actual situation of hardware equipment , Implement a series of interfaces , You can complete the hardware adaptation .

 

EP  It also defines a set of basic computing interfaces  Primitive, be based on  Primitive  The interface is re implemented  Kernel.

comparison  EP  Runtime interface provided ,Primitive  The interface provided is more flexible , Different interfaces are independent of each other , Each interface represents the specific computing power that a certain hardware device can provide .

 

6. Debugging tool stack

 

Add debugging tool stack  OneFlow-Profiler  And  AutoProf.

 

OneFlow-Profiler  A tool used to collect various performance related information in the framework execution process . This tool can count the execution time of operators or system components , Allocation of memory and video memory , At the same time, the input and parameter information corresponding to the operator can be recorded , This information can be used by developers to analyze the most expensive part of the framework implementation , To achieve targeted optimization （

https://github.com/Oneflow-Inc/oneflow/pull/8047

）.

 

 

AutoProf  It's a test  OneFlow  and  PyTorch  Framework for operator performance , by  OneFlow  Provide elegance 、 Detect efficiently  OneFlow API  And  PyTorch API  Whether the method of alignment , Allow users to automatically compare  OneFlow API  And  PyTorch API  Performance results of （

https://github.com/Oneflow-Inc/oneflow/pull/8207

）.

 

7. Handling of abnormal error reports

 

Greatly improve  OneFlow API  Exception handling flow in , perfect  API  Error reporting prompt information under abnormal conditions .

 

8.API file

 

perfect  OneFlow API  The content of the document is 20 The rest , According to the function  API  Document module reconfiguration . Except for general operators  API  Outside , Detailed instructions  OneFlow oneflow.nn.graph、oneflow.embedding、oneflow.autograd  And other modules and environment variables .

see OneFlow v0.8.0 Complete update , Please click on

https://github.com/Oneflow-Inc/oneflow/releases/tag/v0.8.0

Everyone else is watching

• OneFlow Source list ：GDB Compile debugging

• Reading Pathways： The next step forward is OneFlow

• OneFlow The source code parsing ： Automatic inference of operator signature

• Hinton： My 50 years of in-depth study career and Research on mental skills

• LLVM The father of ： Why should we rebuild AI Infrastructure software

• Quantitative model of parallel computing and its application in deep learning engine

• Large model training is difficult ？ Superior efficiency 、 Easy-to-use “ Li Bai ” Here comes the model library

Welcome to download experience  OneFlow v0.8.0  The latest version ：

https://github.com/Oneflow-Inc/oneflow/