Article
ArticleNewsWebsite navigation
Home>Website navigation>AI Big Model>Wudao·Aquila Large Language Model

Wudao·Aquila Large Language Model

Wudao·Aquila Large Language Model

According to an article published today on the WeChat public account of Zhiyuan Research Institute,Zhiyuan Research Institute announced that the Aquila large language model series of Wudao Tianying has been fully upgraded to Aquila2, and added a new heavyweight member with 34 billion parameters (34B). Aquila2-34B achieved a leading overall ranking in 22 evaluation benchmarks. Academia Sinica called it the strongest open source Chinese-English bilingual large model.

In addition to setting a new record for the big model list, ARI has also focused on improving the actual capabilities of important models such as reasoning and generalization, and has achieved a series of results in supporting scenarios such as intelligent agents, code generation, and literature retrieval.

It is worth mentioning that the AI Research Institute has brought a whole family of open source products at once, making innovative training algorithms and practices available simultaneously, including:

1. Comprehensively upgrade the Aquila2 model series: Aquila2-34B/7B basic model, AquilaChat2-34B/7B conversation model, AquilaSQL "text-SQL language" model.

2. The new version of the semantic vector model BGE has been upgraded to fully cover the four major search requirements.

3. FlagScale is an efficient parallel training framework with industry-leading training throughput and GPU utilization.

4. FlagAttention is a high-performance Attention operator set that innovatively supports long text training and Triton language.

Aquila2 model full series open source address:
https://github.com/FlagAI-Open/Aquila2
https://model.baai.ac.cn/
https://huggingface.co/BAAI

AquilaSQL open source repository address:
https://github.com/FlagAI-Open/FlagAI/tree/master/examples/Aquila/Aquila-sql

FlagAttention open source code repository:
https://github.com/FlagOpen/FlagAttention

BGE2 open source address
paper: https://arxiv.org/pdf/2310.07554.pdf
model: https://huggingface.co/BAAI/llm-embedder
repo: https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/llm_embedder

More

  • Pangu Model

    Pangu Big Model 3.0 is an industry-oriented big model, including 5 basic big models in L0, L1 industry big model and L2 scenario model. It is committed to deepening the industry and creating industry big models and capability sets in the fields of finance, government affairs, manufacturing, mining, meteorology, railways, etc. It combines industry knowledge know-how with big model capabilities, reshapes thousands of industries, and becomes an expert assistant for organizations, enterprises and individuals. High-quality data: Federal LLM architecture, jointly create industry data sets Pangu Big Model Development Kit: 5 times faster development of industry big models E2E security compliance system: end-to-end security compliance…

  • iFlytek Spark Cognitive Model

    iFlytek Spark iFlytek's new generation of cognitive intelligence models has cross-domain knowledge and language understanding capabilities, and can understand and execute tasks based on natural conversations. It continues to evolve from massive data and large-scale knowledge, and realizes a closed loop of the entire process from proposing, planning to solving problems.

  • Tencent Hunyuan Large Model

    Tencent Hunyuan Big Model is a large language model independently developed by Tencent. It has over 100 billion parameters and over 2 trillion tokens of pre-trained corpus. It has strong Chinese comprehension and creation capabilities, logical reasoning capabilities, and reliable task execution capabilities. Tencent Hunyuan's product functions Tencent Hunyuan's application scenarios Document scenario: It can provide document creation, text polishing, text review, table formula and chart generation capabilities to improve creation efficiency and enhance the creation experience; Meeting scenario: It can provide in-meeting Q&A, meeting summary, meeting to-do items sorting and other capabilities to simplify meeting operations and improve meeting efficiency; Advertisement...

  • Discussing large language models

    As a member of SenseTime's AGI general artificial intelligence family, it can communicate and interact with people using natural language, and is committed to making AI technology accessible to the general public and becoming a good helper in people's lives and work.

  • Youdao Zhiyun AI Open Platform

    NetEase Youdao is a cloud service platform that provides natural language translation, text recognition OCR and other services as well as industry solutions for developers, enterprises and government agencies. It is committed to providing secure, reliable and efficient cloud services.

  • A Word from the Heart

    Wenxin Yiyan Knowledge Enhanced Large Language Model Baidu's new generation of knowledge enhanced large language model, a new member of the Wenxin large model family, can interact with people, answer questions, assist in creation, and help people obtain information, knowledge and inspiration efficiently and conveniently. 1

  • Thousand Questions on Tongyi

    Alibaba's "Tongyi Qianwen" is a large-scale language model with functions including multi-round dialogue, copywriting, logical reasoning, multi-modal understanding, and multi-language support. It can interact with humans in multiple rounds, integrates multi-modal knowledge understanding, and has very strong copywriting capabilities. It can continue to write novels, write emails, etc., which can greatly increase our work efficiency.

  • HuggingFace

    The AI community builds the future. Build, train, and deploy state-of-the-art models backed by reference open source in machine learning.

❯
Search

  • Scan to open current page

Top