All articles tagged with #hugging face
Chinese AI models, led by DeepSeek R-1 and Qwen, are increasingly powering U.S. companies like Pinterest and Airbnb thanks to open-source access and lower costs, a trend Stanford researchers say has Chinese models catching up or ahead of peers, signaling a shift toward open AI dominance away from heavily proprietary US systems.
Thomas Wolf, co-founder of Hugging Face, argues that current AI models like ChatGPT are unlikely to produce major scientific breakthroughs, as they tend to predict the most likely next word rather than generate novel ideas, contrasting with the contrarian thinking often required for groundbreaking science. He envisions AI as a tool to assist scientists rather than replace the need for true innovation, highlighting existing applications like DeepMind's AlphaFold and startups aiming to push AI further into scientific discovery.
OpenAI has launched two open-weight AI reasoning models, gpt-oss-120b and gpt-oss-20b, available for free download on Hugging Face, marking its return to open sourcing since GPT-2. These models are designed to perform complex queries and can be integrated with more capable closed models if needed. They outperform some open models in benchmarks but tend to hallucinate more. OpenAI emphasizes its commitment to democratizing AI and enabling wider access, releasing the models under a permissive license while withholding training data due to legal concerns.
Hugging Face has started accepting orders for its open-source Reachy Mini desktop robots, available in wireless and wired versions, aimed at AI developers for building and testing AI applications, with a focus on community-driven customization and control.
Hugging Face has launched Reachy Mini, a $299 open-source humanoid robot aimed at democratizing robotics development and challenging traditional high-cost, closed-source industry models. The robot, designed for developers and educators, integrates with the Hugging Face platform, supports multiple programming languages, and emphasizes community-driven innovation, potentially transforming the robotics industry by making physical AI accessible and affordable.
StarCoder 2, an open source code generator developed by Hugging Face in collaboration with ServiceNow and Nvidia, has been released in three variants, offering improved performance and the ability to run on most modern consumer GPUs. The AI model, trained on a larger and more diverse dataset, aims to make coding more efficient for developers. However, concerns about security vulnerabilities and licensing restrictions have been raised, despite claims of improved efficiency and ethical training practices.
Google Cloud's partnership with Hugging Face allows developers to access Google's supercomputers for building, training, and deploying AI models without needing a Google Cloud subscription. Hugging Face, a popular AI model repository, hosts over 350,000 open-sourced models and has received significant investment. The partnership will provide cost-effective access to Google's tensor processing units and GPU supercomputers, benefiting developers using Hugging Face's platform. Google's support for open-source AI ecosystem development is furthered through this collaboration, which will enable Hugging Face users to utilize Google Cloud's AI app-building platform and Kubernetes engine in the first half of 2024.
Researchers at Lasso Security discovered over 1,500 exposed API tokens on the Hugging Face platform, including tokens from tech giants Meta, Microsoft, Google, VMware, and more. These exposed tokens granted write permissions, allowing potential attackers to modify files in account repositories. The researchers were able to gain access to 723 organizations' accounts, including those of Meta, EleutherAI, and BigScience Workshop. If exploited, these tokens could have led to data theft, poisoning of training data, and stealing of models, impacting over 1 million users. The exposed tokens have since been revoked and the vulnerabilities closed.
Hugging Face's two-person team, known as H4, is focused on developing tools and "recipes" to enable the AI community to build AI-powered chatbots similar to ChatGPT. H4 has released open source large language models, including Zephyr-7B-α and Falcon-40B, and relies on a dedicated cluster of over 1,000 Nvidia A100 GPUs for training. While H4 doesn't directly monetize its tools, they contribute to Hugging Face's enterprise-focused offering. The team aims to empower the open AI community by releasing training code and datasets associated with their chat models.
AI startup Hugging Face has raised $235 million in funding, bringing its valuation to $4.5 billion. The investment comes from major tech companies including Google, Amazon, Nvidia, Salesforce, AMD, Intel, IBM, and Qualcomm. Hugging Face provides a platform for AI developers to share code, models, and data sets, making it easier to run open-source AI models. The company's collaborative approach and its focus on developer tools have attracted a large user base, including employees from the investing companies. Hugging Face hosts 500,000 AI models, 250,000 data sets, and has 10,000 paying customers.
NVIDIA and Hugging Face have partnered to bring generative AI supercomputing to developers building large language models (LLMs) and advanced AI applications. The collaboration will provide developers with access to NVIDIA DGX Cloud AI supercomputing within the Hugging Face platform, enabling them to train and customize LLMs with business data for industry-specific applications. Hugging Face will also offer a new service called Training Cluster as a Service, powered by NVIDIA DGX Cloud, to simplify the creation of new and custom generative AI models. The integration is expected to be available in the coming months.
OpenAI's GPT-4 API is not publicly available, but there are platforms that allow users to test it for free, including Forefront AI, Hugging Face, and Microsoft Bing. Poe and Ora.sh also offer limited access to GPT-4. These platforms provide personalized chatbot experiences and the ability to generate high-quality images.
Hugging Face and ServiceNow Research have released StarCoder, a free alternative to code-generating AI systems like GitHub's Copilot. StarCoder was trained on over 80 programming languages and text from GitHub repositories, including documentation and programming notebooks. It integrates with Microsoft's Visual Studio Code code editor and can follow basic instructions and answer questions about code. StarCoder is part of Hugging Face's and ServiceNow's over-600-person BigCode project, launched last year, which aims to develop "state-of-the-art" AI systems for code in an "open and responsible" way.
Hugging Face has released HuggingChat, an open source ChatGPT clone that is available for anyone to use or download. HuggingChat is based on the Open Assistant Conversational AI Model and was trained with the OpenAssistant Conversations Dataset. The dataset is the product of a worldwide crowdsourcing effort by over 13,000 volunteers. HuggingChat is available for use, but it is still in its early stages of development.
Hugging Face has launched an open-source alternative to ChatGPT called HuggingChat, which is essentially a user interface that allows people to interact with an open-source chat assistant dubbed Open Assistant. HuggingChat will soon allow users the ability to plug in new chat models. However, there are licensing issues that need to be worked out before HuggingChat can be used commercially. Hugging Face is iterating quickly on the interface and safety mechanisms and intends to support the next rapidly improving open-source models.