The global large model market has set off a hot boom, and ChatGPT and GPT-4 have appeared one after another, leading the trend! Domestic enterprises responded positively and launched their own large-model products. However, to be comparable to ChatGPT or even comparable to GPT-4, domestic large models still have a long way to go. Today, we take stock of the main models in China and take you to explore this new field!
01-Baidu "Wen Xin Yi Yan"
Access address: https://yiyan.baidu.com/welcome
Free or not: Just submit it and wait for the test!
Baidu's "Wen Xin Yiyan" is a new generation of knowledge-enhanced big language model, which can interact with people, answer questions, assist in creation, and help people obtain information, knowledge and inspiration efficiently and conveniently. Based on Baidu's deep learning platform Feijiao, the model integrates and learns massive data and large-scale knowledge, and has the technical characteristics of knowledge enhancement, retrieval enhancement and dialogue enhancement.
02-Ali "Tongyi Qianqian"
Access address: https://tongyi.aliyun.com/
Free or not: Use the invitation code for the beta!
Alibaba's "Tongyi Qianwen" is a natural language processing model based on the Transformer architecture, developed by Alibaba Natural Language Processing Lab. As one of the largest Chinese pre-training models in China, the model has 100 billion parameters and can realize a variety of natural language processing tasks such as text classification, named entity recognition, and sentiment analysis. In addition, Tongyi Qianwen has strong transfer learning capabilities, which can be applied to various practical application scenarios.
03-Huawei "Pangu"
Access address: https://www.huaweicloud.com/cloudplus/eleventhphase/detail06.html
Free: Invite a closed beta!
Huawei's Pangu series of AI models include NLP models, CV models, and scientific computing models. NLP large models can help enterprises build their own models, efficiently predict results, and improve development efficiency. CV large model is suitable for image classification, object detection, segmentation and other tasks. The big model of scientific computing can complete various scientific computing tasks. The accuracy of the Pangu model in text classification tasks has reached the industry-leading level.
04-SenseTime "New every day"
Access address: https://techday.sensetime.com/
Free: Invite a closed beta!
SenseTime's "Daily New" big model system includes functions such as natural language generation, text generation images, perceptual model annotation, and model research and development. Based on the strategy of "big model + large computing power", breakthroughs have been made in the field of general artificial intelligence. This large model series can improve the efficiency and quality of copywriting, online consultation, and video in seconds.
05-iFLYTEK "Spark Cognition"
Access address: https://xinghuo.xfyun.cn/
Free: Released on the 5.6th!
iFLYTEK's "Spark Cognition" is a large model applied to education, medical care, human-computer interaction and other industries. At the recent AI Big Model Development Forum, iFLYTEK revealed that they will release the "1+N Cognitive Intelligence Big Model" on May 6, where "1" stands for general cognitive intelligence big model algorithm development and efficient training, and "N" represents a dedicated large model version applied to multiple industries. At the same time, they will also demonstrate exemplary application products for "N" scenarios.
06-Kunlun Wanwei "Tiangong"
Access address: https://tiangong.kunlun.com/
Free or not: Apply for a closed beta!
As a large-scale language model, Kunlun Wanwei's "Tiangong" has powerful natural language processing and intelligent interaction capabilities. It can realize a variety of application scenarios such as intelligent question and answer, chat interaction, text generation, etc., and has a rich knowledge reserve, covering science, technology, culture, art, history and other fields. If you need copywriting, knowledge Q&A, code programming, logic deduction, mathematical estimation and other functions, Tiangong will be your right-hand man.
07-Daguan Data "Cao Value"
Access address: http://www.datagrand.com/products/aigc/
Free or not: Apply for a closed beta!
Daguan Data's "Cao Value" is a dedicated domestic ChatGPT model, which aims to realize functions such as automatic text generation, one-click template drafting, intelligent paragraph rewriting, intelligent text review and step-by-step guided writing. As a vertical, dedicated, independent and controllable domestic ChatGPT model, the Cao value model can be built into various business systems of customers to provide special services.
08-Lanzhou Technology "Mencius"
Access address: https://www.langboat.com/technology/mengzi
Free or not: Apply for a closed beta!
Lanzhou Technology's "Mencius" is a large-scale pre-trained language model based on the team's self-developed technology. It can handle multilingual, multimodal data and supports a variety of text understanding and text generation tasks. Whether you need to extract announcements, generate novels, classify emotions, news summaries, Q&A systems, or cross-check graphics and texts, Mencius Model can quickly meet the needs of your different fields and application scenarios.
09-360 "Wisdom Brain"
Access address: https://www.so.com/
Free or not: Apply for a closed beta!
360's "Smart Brain" is a search engine product based on artificial intelligence technology. It uses advanced natural language processing technology to intelligently identify user needs and provide more accurate results through voice interaction and text input. In addition, Brainbrain is deeply integrated with browser, intelligent marketing, soda office, digital assistant and other scenario applications to help you improve user experience and productivity.
10-Institute of Automation, Chinese Academy of Sciences "Zidong Taichu"
Access address: https://www.hiascend.com/zh/software/modelzoo/models/detail/L/4e38d46c0bd211eca5a90242ac110002
Free or not: Apply for a closed beta!
The "Zidong Taichu" launched by the Institute of Automation of the Chinese Academy of Sciences is a super-large model with a scale of 100 billion parameters, which can achieve efficient collaboration between the three modes of vision, text and speech. It shows super performance in tasks such as cross-modal understanding and generation of graphics and text, video understanding and description, and multimodal dialogue virtual human, and has become the current SOTA model in the industry.
11- NetEase "Yuyan"
Access address: (to be announced)
Free or not: Apply for a closed beta!
NetEase's "Yuyan" is a series of Chinese text pre-trained large models. Developed from the earliest GPT2 generative model, the Yuyan series not only supports word games, intelligent NPCs and other applications, but also assists in text creation, music creation, art design and other fields. The Yuyan model has gradually expanded to various modes such as graphics, music, and behavioral sequences, providing users with a richer and more diverse experience.
12- KLCII "Enlightenment 2.0"
Access address: https://www.baai.ac.cn/portal/article/index/cid/49/id/518.html
Free or not: Apply for a closed beta!
KLCII's "Wudao 2.0" is China's first trillion-level model and the world's largest trillion-level model. The parameter scale of more than 1.75 trillion makes it the largest AI pre-training model. Wudao 2.0 can realize the functions of poetry creation, writing couplets, text summaries, character questions and answers, painting and so on. The release of this huge model has had a huge impact on a global scale.
13- Fudan "MOSS"
Access address: https://moss.fastnlp.top/
Free or not: Open source!
Fudan University's open-source "MOSS" is a conversational language model that supports bilingual Chinese and English and multiple plugins. It is based on the General Language Model (GLM) architecture and has 16 billion parameters. Through a large number of training in Chinese and English and code, MOSS has a variety of functions such as logical reasoning, image generation, Chinese context understanding, and code programming to meet various practical needs.
14- Tsinghua "ChatGLM-6B"
Access address: https://chatglm.cn/login?redirect=%2F
Free or not: Open source!
Tsinghua University's "ChatGLM-6B" is an open-source conversational language model that supports both Chinese and English. Based on the model of 6.2 billion parameters, it supports web, command line, and API deployment. The ChatGLM-6B has multi-turn dialogue capabilities, being able to answer questions and provide answers that match human preferences. It offers excellent performance and results in inference and generation.
15- Zhihu "Zhihai Chart AI"
Access address: (to be announced)
Free: Not open for closed beta!
The "Zhihai Chart AI" jointly developed by Zhihu and Face Wall Intelligence has begun to be applied to the Zhihu hot list. "Zhihai Chart AI" uses strong language understanding capabilities to organize and aggregate the answers to questions on the Zhihu hot list, and present them to users in the form of a synopsis to answer questions intuitively and clearly. The closed beta of this feature will be opened soon, providing users with a better service experience.
16- Supersymmetry "BBT-2"
Access address: https://bbt.ssymmetry.com/
Free: Request a trial!
The "BBT-2" launched by supersymmetry is a large-scale pre-trained model based on the GPT Decoder-only architecture. They open-sourced the BBT-1-0.2B model and newly introduced models such as BBT-1-1B, BBT-2-12B-Text, and BBT-2.5-13B-Text. These models cover multiple high-quality data sources, including books, encyclopedias, papers, novels, news, policy documents, and more. BBT-2-12B-Text and BBT-2.5-13B-Text have stronger expressive and predictive abilities.
17-Kingsoft Office "WPS AI"
Access address: (to be announced)
Free or not: Closed beta stage!
The "WPS AI" launched by Kingsoft Office is the first large-model application in the field of collaborative office. Equipped on light document products, it provides functions such as text generation, multi-round dialogue, and editing and rewriting. In the future, it will be embedded in Kingsoft office full-stack products to help users achieve more efficient results in content writing and editing.
18- Go out and ask "sequence monkeys"
Access address: https://openapi.mobvoi.com/largemodel-introduce
Free or not: Apply for a closed beta!
Mobvoi released "Sequence Monkey" is a very large-scale language model with the characteristics of long sequence, multimodality, single model, and big data. It is suitable for question answering systems, natural language processing, machine translation, text summarization, and more. Sequence monkey improves production efficiency and data processing capabilities by efficiently cooperating multiple modal data.
The above is a selection of large domestic models, which have excellent performance and functions in different fields and application scenarios. Visit the corresponding links, apply for the closed beta, and experience these powerful and advanced models to help you get information more efficiently, solve problems, and bring new possibilities to your work!