OpenBMB’s Events of 2022

OpenBMB’s Events of 2022. We build BMSystem for training, tuning… | by OpenBMB | Medium

Sign up

Get app

Sign up

OpenBMB’s Events of 2022

4 min read

Jan 6, 2023

--

Share

We build BMSystem for training, tuning and inference of big model.

We have built the capability system of big model step by step. We launched a series of full-process acceleration tools of big model, which includes three suites for training, tuning, and inference. The training suite includes the data collector BMData and the training engine BMTrain. The tuning suite includes the prompt learning tool OpenPrompt, the delta tuning tool OpenDelta, and Delta Center, the delta object sharing center Delta Center. The inference suite includes the high efficiency compression tool BMCook, and BMInf, a tool for efficient inference. OpenPrompt has gotten 2.2k+ stars on GitHub since its release, and other tools also got 1.3k+ stars altogether.

Press enter or click to view image in full size

We launch CPM-Live, Ant finished and Bee started.

We launched CPM-Live, a live training for open-source big models, and the proposal was released on May 26th. The training of the first model CPM-Ant was officially launched on May 29th, which took 68 days and was completed on August 5, and the report was finally released successfully on September 16th. CPM-Ant actualized five characteristics: efficient computation, excellent performance, economical deployment, convenient use and open source.The training of the second model CPM-Bee started on October 13th, which added some new features such as task mode enhancement, multilingual fusion, and complex structure processing. The training is almost complete, so stay tuned!

Press enter or click to view image in full size

Our scientific research has yielded fruitful results, paving the way for big model applications.

In February 2022, our team released KV-PLM, a big model in the biomedical field, which paper was selected into Editors’ Highlights by Nature Communications. In March, the corresponding papers of BMInf and OpenPrompt were accepted by the ACL 2022 Demo. BMInf supports inference and tuning of big models at a greatly lowered computational cost, and OpenPrompt provides a unified interface for prompt learning template language. In May, the open source toolkit tool OpenPrompt won the Best Presentation Paper Award of the top conference ACL. On the basis of the above scientific research achievements, we continue to improve and apply big model tools to promote the practical transformation of academic achievements.

Press enter or click to view image in full size

Press enter or click to view image in full size

We create a big model community and build the open source ecology.

We have built a multi-integrated open source ecosystem of big models. All open source projects of OpenBMB on GitHub have provided developers with a good experience, which have earned 3.5k+ star. We provide relevant information on many Chinese platforms such as WeChat, Blibli, Zhihu and the BAAI community. The total number of our followers across all platforms is more than 6k, and nearly 2,000 interested people have joined our communication community to discuss the big model issues in depth.

Get OpenBMB’s stories in your inbox

Join Medium for free to get updates from this writer.

Subscribe

Subscribe

Remember me for faster sign in

We have also released English information on international platforms, including Twitter, Medium, Hacker News. In the future, we will strive to build a global big-model communication community.

We have released open courses and popularize big model.

Our big model related courses are very popular in China. On July 29th, we released an 18-hour open course, including 9 courses, on Blibli. It is the first systematic big-model open course in China. So far, the broadcast volume has reached over 3w, and the course has received high praise.On October 20th, our course also got online on Zhihu, and the response is very enthusiastic. Up to now, the total number of learning times has been 43w+.

Since September 29th, we have launched the “Paper Speed Reading” column on both our WeChat official account and Blibli, which focuses on leading followers to quickly master cutting-edge classic papers in 10min with mind maps. So far, we have read papers of EMNLP, ACL and ICLR in succession, with a total video play of 1w+.

We have participated in both online and offline activities.

Based on the precipitation of the first three quarters, we participated in various communication activities in the fourth quarter to share the open source achievements with the public, and also to exchange experience with industry colleagues.

On October 29th, we participated in the offline Open source market of Beijing branch venue of COSCon’22. The atmosphere was very warm. We participated in the “New Generation AI Infrastructure and Applications” forum of DataFunSummit 2022 on November 19th. Then on December 10th, we jointly held the “2022 Big Model Innovation Forum · Training Camp” with BAAI and PARATERA. Finally, on December 22nd, we participated in the 4th lecture of “Big Model Series Live Class” of Zhidx Course.

Thanks to your sincere interest and participation in OpenBMB, we have ignited the solitary spark of the big model, and we will continue to see how it evolves along the way.

Happy New Year to you all!

Written by OpenBMB

OpenBMB, known as Open Lab for Big Model Base, aims to build a large-scale pre-trained language model library and related tools

No responses yet

More from OpenBMB

·

Nov 17, 2022

OpenBMB: Big Models for Everyone

In recent years, as the pre-trained language model technology has triggered a performance revolution in artificial intelligence, maturity…

·

Nov 16, 2023

ModelBest Adopts the LLM-Based AI Agents to Launch ChatDev, its First SaaS Product

Breaking news! Here comes an AI-native app developed using the LLM-based AI agents.

·

Jul 15, 2023

Speak and draw! VisCPM：SOTA Open-source Chinese Multimodal Large Model

VisCPM is a series of open-source large multimodal models, supporting multimodal conversational and and text-to-image generation…

·

Jun 17, 2023

Official Tutorial — How to Fine-tune CPM-Bee on Basic Tasks

A formal and detailed tutorial of how to fine-tune the foundation model CPM-Bee and its data formats…

See all from OpenBMB

Recommended from Medium

In

by

Dr. Patricia Schmidt 🧠

·

Jan 14…