Technology peripherals
AI
TPC Alliance established: Targeting AI models with more than one trillion parameters to promote scientific discovery
TPC Alliance established: Targeting AI models with more than one trillion parameters to promote scientific discovery
According to news on November 16, leading scientific research institutions in the industry, the US National Supercomputing Center and many leading companies in the AI field, recently jointly formed the Trillion Parameter Consortium (TPC).
Generated by DALL-E 3
According to reports, this site learned that the TPC Alliance consists of laboratories, scientific research institutions, and academia around the world and scientists from industry to jointly advance artificial intelligence models for scientific discovery, with a special focus on giant models with a trillion parameters or more
The TPC consortium is currently working to develop scalable Model architecture and training strategies, while organizing and collating scientific data for model training to optimize the application of AI libraries on current and future exascale computing platforms
TPC aims to create an open A community of researchers developing large-scale generative AI models for scientific and engineering problems, in particular, will launch joint projects to avoid duplication of work and share methods, approaches, tools, knowledge and workflows. In this way, the consortium hopes to maximize the impact of these projects on the broader artificial intelligence and scientific communities.
TPC’s goal is to build a global network of resources, data and expertise. Since its inception, the consortium has established multiple working groups aimed at addressing the complexities of building large-scale artificial intelligence models. The exascale computing resources required for training will be provided by the U.S. Department of Energy (DOE). Available from several national laboratories in , as well as several TPC founding partners in Japan, Europe, and other countries. Even with these resources, training can take months.
Rick Stevens, associate director for computational, environmental and life sciences at the U.S. Department of Energy's Argonne National Laboratory and professor of computer science at the University of Chicago, said: "In our laboratory and with partner institutions around the world In the process of collaboration, our team is beginning to develop a series of cutting-edge artificial intelligence models for scientific research and is preparing to use large amounts of previously untapped scientific data for training."
The above is the detailed content of TPC Alliance established: Targeting AI models with more than one trillion parameters to promote scientific discovery. For more information, please follow other related articles on the PHP Chinese website!
Hot AI Tools
Undresser.AI Undress
AI-powered app for creating realistic nude photos
AI Clothes Remover
Online AI tool for removing clothes from photos.
Undress AI Tool
Undress images for free
Clothoff.io
AI clothes remover
Video Face Swap
Swap faces in any video effortlessly with our completely free AI face swap tool!
Hot Article
Hot Tools
Notepad++7.3.1
Easy-to-use and free code editor
SublimeText3 Chinese version
Chinese version, very easy to use
Zend Studio 13.0.1
Powerful PHP integrated development environment
Dreamweaver CS6
Visual web development tools
SublimeText3 Mac version
God-level code editing software (SublimeText3)
Hot Topics
1386
52
Is Flash Attention stable? Meta and Harvard found that their model weight deviations fluctuated by orders of magnitude
May 30, 2024 pm 01:24 PM
MetaFAIR teamed up with Harvard to provide a new research framework for optimizing the data bias generated when large-scale machine learning is performed. It is known that the training of large language models often takes months and uses hundreds or even thousands of GPUs. Taking the LLaMA270B model as an example, its training requires a total of 1,720,320 GPU hours. Training large models presents unique systemic challenges due to the scale and complexity of these workloads. Recently, many institutions have reported instability in the training process when training SOTA generative AI models. They usually appear in the form of loss spikes. For example, Google's PaLM model experienced up to 20 loss spikes during the training process. Numerical bias is the root cause of this training inaccuracy,
TPC Alliance established: Targeting AI models with more than one trillion parameters to promote scientific discovery
Nov 18, 2023 pm 07:29 PM
According to news on November 16, leading scientific research institutions in the industry, the US National Supercomputing Center and many leading companies in the AI field have recently jointly established the Trillion Parameter Consortium (TPC). Generated by DALL-E3 According to reports, this site has learned that the TPC Alliance is composed of scientists from laboratories, scientific research institutions, academia and industry around the world. It aims to jointly promote artificial intelligence models for scientific discovery, and pays special attention to having a The TPC Consortium is currently working to develop scalable model architectures and training strategies for mega-models with one trillion parameters or more, while organizing and curating the scientific data used for model training to optimize AI libraries for current and future exascale applications. level computing platform
Microsoft launches XOT technology to enhance the reasoning capabilities of language models
Nov 17, 2023 pm 05:45 PM
According to news on November 15, Microsoft recently launched a method called "Everything of Thought" (XOT), inspired by Google DeepMind's AlphaZero, which uses compact neural networks to enhance the reasoning capabilities of AI models. Microsoft collaborated with Georgia Institute of Technology and East China Normal University to develop this algorithm, which integrates reinforcement learning and Monte Carlo Tree Search (MCTS) capabilities to further improve the effectiveness of problem solving in complex decision-making environments. Note from this site: The Microsoft research team stated that the XOT method can expand the language model on unfamiliar problems. In Gameof24, 8-Puzzle and P
Google's DeepMind has developed the RoboCat AI model, which can control a variety of robots to perform a series of tasks
Jun 26, 2023 pm 04:07 PM
According to news on June 26, DeepMind, a subsidiary of Google, said that the company has developed an artificial intelligence model called RoboCat that can control different robot arms to perform a series of tasks. This alone isn't particularly novel, but DeepMind claims that this model is the first to be able to solve and adapt to a variety of tasks, and to do so using different, real-world robots. RoboCat is inspired by another DeepMind AI model, Gato, which can analyze and process text, images and events. RoboCat's training data includes images and motion data of simulated and real robots, which come from other robot control models in the virtual environment, human-controlled robots
Databricks releases AI model SDK for big data analysis platform Spark: one-click generation of SQL and FySpark language chart code
Jul 17, 2023 pm 05:53 PM
According to news on July 10, Databricks recently released the AI model SDK used by the big data analysis platform Spark. When developers write code, they can give instructions in English, and the compiler will convert the English instructions into PySpark or SQL language codes to improve developers' efficiency. ▲Image source Databricks website It is reported that Spark is an open source big data analysis tool that is downloaded more than 1 billion times a year and is used in 208 countries and regions around the world. ▲Image source Databricks website Databricks said that Microsoft’s AI code assistant GitHubCopilot is powerful, but the threshold for use is also quite high. Databricks’ SDK is relatively more universal and easier to use.
The 'FunSearch” training method announced by Google DeepMind: enables AI models to solve complex discrete mathematical problems
Dec 17, 2023 pm 08:15 PM
According to news on December 15, Google DeepMind recently announced a model training method called "FunSearch", which claims to be able to calculate a series of "involving the fields of mathematics and computer science" including "upper-level problems" and "boxing problems". complex issues." The content that needs to be rewritten is: ▲Source: Google DeepMind (hereinafter referred to as DeepMind) It is reported that the FunSearch model training method mainly introduces an "Evaluator" system for the AI model, and the AI model outputs a series of "creative problem-solving methods" ", and the "evaluator" is responsible for evaluating the problem-solving methods output by the model. After repeated iterations, an AI model with stronger mathematical capabilities can be trained. Google's DeepM
Microsoft launches LLaVA-Med AI model to analyze medical pathology cases
Jun 15, 2023 pm 03:06 PM
According to news on June 14, Microsoft researchers recently demonstrated the LLaVA-Med model, which is mainly used for biomedical research and can infer the pathological conditions of patients based on CT and X-ray pictures. It is reported that Microsoft researchers have cooperated with a group of hospitals and obtained a large data set corresponding to biomedical image text to train a multi-modal AI model. This data set includes chest X-ray, MRI, histology, pathology and CT images, etc., with relatively comprehensive coverage. ▲Picture source Microsoft Microsoft uses GPT-4, based on VisionTransformer and Vicuna language model, to train LLaVA-Med on eight Nvidia A100 GPUs, which contains "all pre-analysis information for each image",
Microsoft releases latest AI terms of service: Reverse engineering and other activities are prohibited
Aug 16, 2023 pm 05:53 PM
Microsoft announced its AI service terms on August 16 and announced that the terms will take effect on September 30. The main content of this update is for generative AI, especially content related to the use of relevant users and responsible development practices. Microsoft emphasizes that the official will not retain the conversation records of users chatting with Bing, nor will these chat data be used. The five key policy points used to train the AI model for Bing Enterprise Chat cover multiple areas, including prohibiting users from attempting to reverse engineer the AI model to prevent revealing underlying components; prohibiting data extraction through methods such as web scraping unless explicitly allowed; An important clause restricts users from using AI data to create or enhance other AI services. The following is a clause added by Microsoft.
