夜深了，注意休息，愿你今夜好梦。
Since its establishment in 2023, DeepSeek (Hangzhou DeepSeek) has rapidly launched open source big models such as DeepSeekCoder, DeepSeek-V3 and DeepSeek-R1. Its innovative MoE architecture dramatically reduces inference costs, and its products have landed on the NVIDIA NIM platform and gone live on the National Supercomputing Internet. This article combs through the company's development timeline, technological breakthroughs and industry cooperation.
Tencent Hybrid is Tencent's chain-wide self-research big model platform, providing AI services such as text, image, video, 3D asset generation and logical reasoning. The platform is enterprise-oriented through Tencent Cloud APIs, supporting multi-scenario applications such as content creation, enterprise services, intelligent assistants, etc. It excels in Chinese semantic understanding and multi-modal interaction.
Meta Llama official website is the core platform for obtaining the open source large language model LLaMA series. The website provides LLaMA 1/2/3 and other versions of the model's technical documentation, download resources and API guide, covering text generation, multimodal understanding, code generation and other application scenarios, to facilitate developers and researchers to quickly get started and deployment.
Beijing Wisdom Source Artificial Intelligence Institute (BAAI) is a non-profit organization focusing on basic artificial intelligence research. Its official website is the core platform for accessing its cutting-edge research results (e.g., big models such as Wudao and Tele-FLM), open-source tools (e.g., BGE Vector Model, FlagOpen platform), and participating in activities such as the Zhiyuan Conference and the Zhiyuan Community. The website provides a wealth of resources and collaboration opportunities for researchers, developers and industry.
Tongyi Wanxiang is an AI creative platform launched by Aliyun, providing core functions such as text-to-diagram, diagram-to-video, and image style migration. The platform supports a variety of drawing styles and video ratios, has a built-in structured cue thesaurus, and features material management and project set functions. Users can get free inspiration value every day, and developers can also access the service through the API.
Gitee AI is a one-stop AI modeling service platform under Open Source China. The platform brings together mainstream models such as DeepSeek, Qwen2.5, Kolors, etc., and provides online experience, inference training, Serverless API deployment and community support. Targeting developers, researchers and enterprise users, it aims to simplify the AI application development process and lower the threshold of computing power usage.
Tianyang Xiaobai is a large model application development platform for developers and general users. It allows users to configure cue words, adjust the engine and quickly create personalized AI applications without programming through a visual interface and rich components. The platform has a built-in self-developed large language model and a variety of parameter models, and provides an accurate semantic search engine that supports efficient retrieval of documents such as PDF, Word, etc., aiming to provide accurate answers based on facts and reduce the "illusion" of large models.
InternLM is an open source AI platform operated by Shanghai Artificial Intelligence Laboratory, providing a series of products such as large language models, cloud development environment, multimodal tools and research assistants. The site integrates the whole process of tools and resources from model training, algorithm development to practical applications, which is suitable for developers, researchers and enterprise users to build AI solutions.
GPT-OSS AI platform is based on OpenAI open source models and provides two model options, GPT-OSS 120B and 20B, focusing on fast response and deep inference respectively. The platform features enterprise-grade security standards, fast global access, and instant use without waiting, making it suitable for a variety of professional scenarios such as content creation, programming, and data analysis.
K2 is an open-source large language model designed specifically for the geosciences. It is based on the LLaMA architecture and is first pre-trained in the domain on a large amount of geoscience literature (including open access papers and Wikipedia), and then fine-tuned using the knowledge-intensive instruction data GeoSignal. In the GeoBench benchmark evaluation consisting of NPEE and AP Geology, Geography, and Environmental Science tests, K2 outperforms similar parameter-scaled baseline models on a number of objective and subjective tasks. The project will open source the associated code and data.