Recently, Huggingface, the world's largest open source AI community, released the latest Open LLM Leaderboard. The results show that all ten of the top-ranked open source large models are derivative models that have been fine-tuned based on Alibaba Tongyi Qwen's (Qwen) open-source model. This achievement marks Qwen's dominant position in the open-source AI field and further enhances its global influence. The Open LLM Leaderboard is widely...

Recently, cybersecurity researchers discovered two malicious machine learning models that were uploaded quietly to the renowned machine learning platform HuggingFace. These models utilized a novel technique, successfully bypassing security detection through 'corrupted' pickle files, raising concerns. Karlo Zanki, a researcher at ReversingLabs, pointed out that the beginning of the pickle files extracted from these PyTorch format archives suggests that

Hugging Face, in collaboration with Physical Intelligence, has launched the groundbreaking robotic foundation model Pi0. This is the first open-source model capable of directly transforming natural language commands into robotic actions, marking a new era in robotics technology. The Pi0 model has been trained on seven different robotic platforms and has mastered 68 unique tasks, enabling it to perform complex operations ranging from folding clothes to tidying up tables. The model utilizes innovative flow matching technology to generate smooth real-time actions at a frequency of 50Hz.

Recently, Hugging Face partnered with Physical Intelligence to launch 'Pi0' (Pi-Zero), the first foundational model that directly converts natural language commands into physical actions. This innovative release has garnered widespread attention, with Hugging Face's Chief Research Scientist Remi Cadene announcing on social media, 'Pi0 is the most advanced visual-language action model that can translate natural language commands into autonomous behaviors.'

Recently, IBM announced the launch of version 3.1 of its open source language model Granite, which brings many important enhancements. The new version of the model has been redesigned with a denser architecture, capable of handling up to 128,000 tokens at once. This update signifies a substantial improvement in Granite's ability to process complex text and tasks. The Granite 3.1 model has been trained on datasets in 12 languages and 116 programming languages, processing a total of 12 trillion tokens.

Beijing TuSimple Technology Co., Ltd. officially launched its first 'Image-to-Video' large model - 'Ruyi' on December 17, 2024, and has made the Ruyi-Mini-7B version open source for users to download from the huggingface platform. TuSimple was founded in 2015 and is headquartered in San Diego, California, focusing on the application of AI technology across various industries, including animation gaming and transportation.

Recently, the social media platform Bluesky faced a significant data scraping incident. A machine learning librarian, Daniel van Strien, scraped over one million public user posts from Bluesky's API and uploaded this data to the AI company Hugging Face. The dataset includes users' Decentralized Identifiers (DIDs) and a range of features to search for specific user content.

Teuken-7B, a language model with 7 billion parameters, has now been launched on Hugging Face, supporting all 24 official EU languages. This model was developed by the EU OpenGPT-X research project and is available as an open-source project for users. Unlike most AI language models that are English-centric, Teuken-7B was built from the ground up, with approximately half of its training data sourced from non-English European languages. Image source note: The image is AI-generated.

Microsoft's recently launched screen content parsing tool OmniParser has jumped to the top of the Most Popular Models on the artificial intelligence open-source platform HuggingFace this week. According to HuggingFace co-founder and CEO Clem Delangue, this is the first parsing tool to achieve this honor in the field. OmniParser is primarily used to convert screenshots into structured data, aiding other systems in better understanding and processing graphical user interfaces. The tool utilizes a collaborative approach with multiple models: YO

Meta recently announced that its MobileLLM model is now open to researchers, allowing users to download and use these models for free on the Hugging Face platform, fully integrated with the Transformers library. This initiative aims to promote research and development of large language models (LLMs) on mobile devices, providing a broader range of tools and resources for developers and academia. MobileLLM represents a significant advancement for Meta in the field of mobile AI.

At the recent Gemma Developer Day held in Tokyo, Google officially launched its new Japanese version of the Gemma AI model. This model's performance rivals that of GPT-3.5, but it has a parameter count of only 2 billion, making it very compact and suitable for operation on mobile devices. The newly released Gemma model excels in Japanese processing while also maintaining its capabilities in English. This is particularly important for small models, as they may face 'catastrophic forgetting' when fine-tuning for new languages.

Recently, ByteDance officially addressed rumors regarding an 'intern disrupting large model training'. The company confirmed that an intern did maliciously interfere with the model training tasks of the commercialization technology team's research project. However, this action did not affect the company's official commercial projects and online business, nor did it involve other large model businesses of ByteDance. Moreover, ByteDance pointed out that claims circulating online about 'involving over 8,000 nodes and causing losses of over ten million dollars' are seriously exaggerated. According to public reports, the intern was part of ByteDance's commercialization technology team.