2026-06-12 15:56
Introducing **olmo-eval**, an innovative evaluation workbench designed for the model development loop. This tool allows developers to evaluate their LLMs repeatedly during the building process. It adapts to changes in data, architecture, and hyperparameters, ensuring that each model checkpoint is effectively assessed. Unlike traditional evaluation tools, olmo-eval is tailored for dynamic models and real-world conditions. Discover more: π» [GitHub Link](https://github.com/allenai/olmo-eval)...
Source: Hugging Face Blog
2026-06-11 00:00
In the latest article on profiling in PyTorch, the focus shifts from basic operations to using `nn.Linear` to create a Multilayer Perceptron (MLP). This transition highlights key concepts such as the CPU dispatch chain and profiling traces. The post includes scripts to illustrate these points: 02_linear.py, 03_simple_mlp.py, and 03_kernels_mlp.py. For those diving deeper into PyTorch, this is a valuable resource! ππ» #PyTorch #MachineLearning #DeepLearning #MLP #DataScience
Source: Hugging Face Blog
2026-06-09 19:38
Exploring the capabilities of voice agents, a recent article evaluates how Automatic Speech Recognition (ASR) systems perform with bilingual and code-switched speech. The study highlights the challenges these systems face in accurately understanding mixed-language conversations. Findings suggest a need for ongoing improvements to better serve bilingual customers. π€π£οΈπ #VoiceTechnology #BilingualCommunication #ASR #CodeSwitching #Innovation
Source: Hugging Face Blog
2026-06-09 15:56
π Exciting news in the AI development space! Cohere has launched North Mini Code, a 30B-parameter Mixture-of-Experts model with 3B active parameters. This model is tailored for agentic software engineering tasks and is available on Hugging Face under the Apache 2.0 license. North Mini Code excels in complex coding workflows and high-quality code generation, outperforming other models of similar size in benchmarks. Check it out! π»β¨ #Cohere #AI #SoftwareEngineering #MachineLearning #Coding
Source: Hugging Face Blog
2026-06-09 10:46
Discover how an innovative agent created a 3D gallery of Paris by integrating two Hugging Face Spaces. This project showcases the potential of AI technology in building immersive experiences. The article delves into the technical process and tools used to achieve this unique virtual environment. Explore the future of digital galleries! ππ¨ #HuggingFace #3DArt #AIDesign #VirtualReality #Innovation
Source: Hugging Face Blog
2026-06-09 00:00
π Migrating your GitHub CI to Hugging Face Jobs can enhance efficiency and capabilities. Many projects rely on GitHub-hosted runners, but they come with limitations, like slow performance and lack of GPU access. Trackio faced these challenges and transitioned to using Hugging Face Jobs for a faster and more robust CI process. This article provides a step-by-step guide to help you replicate this setup for your own GitHub repository, including CLI and browser-based instructions. #GitHubCI...
Source: Hugging Face Blog
2026-06-08 13:10
In the latest article from the Build Small Hackathon, the author reflects on a unique experiment involving a fictional bank run, "Run on Oona's Hoard." The scenario illustrated how small model roles and budgets can lead to unexpected market behaviors. Initially, a surge in supply caused honey prices to plummet. However, after rebuilding the model, the crash no longer occurred, prompting insights into the complexities of agent-based systems. #EmergentBehavior #MarketDynamics #Innovation...
Source: Hugging Face Blog
2026-06-08 11:46
Introducing the Pakistan Notice Helper, an AI tool designed to enhance safety for users in Pakistan. π΅π° This tool assists in analyzing suspicious messages related to banks, utilities, and more, helping users decide whether to click links or share personal information. While it doesn't verify the authenticity of messages, it aims to empower individuals to make informed choices. #AISafety #CyberSecurity #PakistanTech #Innovation #CommunitySupport
Source: Hugging Face Blog
2026-06-08 00:00
π Exciting developments in the AI community! OpenEnv, a tool for creating agentic execution environments, is now more open than ever. It will be coordinated by a new committee featuring major players like Meta-PyTorch, Nvidia, and Hugging Face. This initiative aims to enhance the training of agents in an open-source manner. Check out the project at huggingface/OpenEnv! π #OpenSource #AI #OpenEnv #AgenticRL #MachineLearning
Source: Hugging Face Blog
2026-06-07 19:44
π Exploring the journey of the "Amazing Digital Dentures" project, inspired by the animated show "The Amazing Digital Circus." The concept started as a digital pet designed to enhance productivity through gamified adventures. However, the project faced challenges when attempts to create functioning games with the AI model fell short. Initial strategies, including long prompts and skill cards, did not yield the desired results. Adjustments to the model's context window also proved ineffective....
Source: Hugging Face Blog
2026-06-07 11:38
OpenAI's Codex challenge offers $10,000 in cash and ChatGPT Pro subscriptions for the best projects. π Participants are encouraged to build their Space using Codex, push the code to a public GitHub repo, and submit the link in their README. There have been some questions regarding voucher activation for both Codex and modal, with some issues resolved. Stay tuned for updates! ππ» #OpenAI #CodexChallenge #TechInnovation #CodingCommunity #GitHub
Source: Hugging Face Blog
2026-06-06 19:02
π Discover the insights from the latest Build Small Hackathon report! The article explores the evolution of the game "Thousand Token Wood," transitioning from a passive experience to an interactive one. Players take on the role of a financier, influencing a dynamic economy with various strategies. In this version, each woodland creature operates using distinct models from different labs, adding complexity to gameplay. #GameDevelopment #Finance #EmergingEconomies #Innovation #ThousandTokenWood
Source: Hugging Face Blog
2026-06-06 15:36
Job hunting as a new graduate can be overwhelming. π Many spend hours sifting through job postings and submitting applications, often feeling fatigued by the process. As time passes, some may find themselves applying to roles they wouldn't typically consider, simply to keep up the momentum. This experience highlights the challenges faced during a job search. #JobSearch #NewGrad #CareerAdvice #JobHunting #Employment
Source: Hugging Face Blog
2026-06-05 22:18
π Exciting insights from the "Thousand Token Wood" project! This report details a tiny economy created during the Build Small Hackathon, featuring five agents trading goods. The model illustrates trading dynamics like bubbles and wealth gaps using a 3-billion-parameter system. Key takeaways include the reliability of the model as a format generator and the importance of designed scarcity in emergent systems. Check out the full report for more details! π²π» #Economy #TechInnovation #AIModels...
Source: Hugging Face Blog
2026-06-04 18:57
π Exciting advancements in AI content safety! NVIDIA has launched Nemotron 3.5, enhancing its content safety model with customizable multimodal and multilingual features. This update integrates various input types and supports enterprise policy enforcement in one call. The article details the improvements and design choices made, alongside integration tips for production safety pipelines. #AI #ContentSafety #NVIDIA #TechInnovation #EnterpriseAI
Source: Hugging Face Blog
2026-06-04 12:59
π Introducing NVIDIA Nemotron 3.5 ASR! This advanced speech-to-text model features 600M parameters and can transcribe in 40 language-locales from a single checkpoint. It includes built-in punctuation and capitalization for real-time accuracy. π Validated by benchmarks, it ranks 2nd in latency, achieving a final transcript just 0.07 seconds after speech ends. #NVIDIA #ASR #SpeechToText #Technology #Innovation
Source: Hugging Face Blog
2026-06-04 12:24
π Exciting updates in the world of data! The EVA-Bench Data 2.0 has introduced a comprehensive platform featuring 3 domains, 121 tools, and 213 scenarios. This enhanced version aims to support enterprises in data management and analysis. Stay informed and leverage these resources to improve your data strategies! #DataManagement #EVA_Bench #EnterpriseTools #DataAnalysis #TechUpdates
Source: Hugging Face Blog
2026-06-04 11:24
π New advancements in large-scale LLM development focus on the quality of data rather than just quantity. The article discusses task-seeded synthetic Q&A generation for Nemotron pretraining. This method enhances model training by providing structured examples that address specific information needs. In a recent experiment, improvements were noted in several areas, including MMLU-Pro and commonsense understanding. Explore how this innovative approach could impact future AI models! π€π #AI...
Source: Hugging Face Blog
2026-06-04 00:00
π The hf CLI is the official command-line tool for the Hugging Face Hub, allowing users to perform various tasks like downloading models and managing repositories directly from the terminal. Recently, it has been optimized for both human users and coding agents like Claude Code and Codex. Benchmark tests show that for complex tasks, the hf CLI is significantly more efficient, using up to 6Γ fewer tokens compared to traditional methods. #HuggingFace #AI #CommandLine #TechUpdate #Efficiency
Source: Hugging Face Blog
2026-06-03 12:55
π Direct Preference Optimization moves beyond traditional chatbots, exploring advanced techniques for user engagement. The article discusses how these methods enhance personalization in digital interactions, offering more tailored experiences for users. It highlights the importance of understanding user preferences to improve satisfaction and drive better outcomes. #UserEngagement #DigitalInnovation #Personalization #TechTrends
Source: Hugging Face Blog
2026-06-03 00:00
π Exciting news for Reachy Mini users! The Reachy Mini conversation app can now utilize tools from public Hugging Face Spaces via MCP. This means you can enhance your robotβs capabilities, such as checking the weather or searching the web, without altering the app's code. Adding tools is simple with just one command, and you can even publish your own tools for others! #ReachyMini #Robotics #AI #HuggingFace #Innovation
Source: Hugging Face Blog
2026-06-02 14:13
π Exciting updates on Holo3.1! This new release builds on the success of Holo3, which has seen widespread adoption in various workflows. Users now demand seamless integration across desktop and mobile environments. Holo3.1 enhances performance with improved robustness for web, desktop, and mobile applications. The update introduces quantized checkpoints for local inference, optimizing user experience. Discover how Holo3.1 is paving the way for universal computer-use agents! #Holo3...
Source: Hugging Face Blog
2026-06-01 15:45
π JetBrains has unveiled Mellum2, a new 12 billion parameter mixture-of-experts model. This advanced model aims to enhance performance in various AI applications. Mellum2 utilizes a unique approach that enables efficient processing and improved decision-making capabilities. Stay tuned for more updates on AI advancements! π€β¨ #AI #JetBrains #Mellum2 #MachineLearning #Innovation
Source: Hugging Face Blog
2026-06-01 13:51
π Scalable AI adoption in enterprises hinges on the concept of agent logic, according to a recent article. AI agents have the potential to transform industries, but they require intelligent guidance to ensure high quality and cost-effectiveness. This is crucial for building user trust. Understanding enterprise workflows is key, as many AI pilots have faced challenges. #EnterpriseAI #AgentLogic #AIFuture #Innovation #TechTrends
Source: Hugging Face Blog
2026-06-01 04:44
NVIDIA has introduced Cosmos 3, marking a significant advancement in physical AI. This new open omni-model enhances reasoning and action capabilities in AI systems. It aims to improve the efficiency and effectiveness of AI applications across various industries. Stay tuned as this technology evolves! π€π‘π #NVIDIA #AI #TechInnovation #Cosmos3 #MachineLearning
Source: Hugging Face Blog
2026-05-29 00:00
Unlock the power of profiling in PyTorch! π This beginner's guide introduces the torch.profiler tool, essential for optimizing your models. The article addresses the challenges of interpreting dense profiler traces and aims to simplify the process. In Part 1, you'll learn to analyze a basic operation: matrix multiplication with bias. Future parts will expand on this foundation, exploring more complex structures. Stay tuned for a deep dive into optimization! π #PyTorch #MachineLearning...
Source: Hugging Face Blog
2026-05-27 17:20
π Exciting news in AI benchmarking! Artificial Analysis and IBM Research have introduced ITBench-AA, focusing on agentic enterprise IT tasks. The first benchmark evaluates performance on Site Reliability Engineering tasks, with frontier models scoring below 50%. The benchmark assesses Kubernetes incident response, requiring models to analyze logs and trace dependencies in complex systems. Future expansions will include Financial Operations and Chief Information Security Officer tasks. #AI...
Source: Hugging Face Blog
2026-05-27 00:00
π Exciting news for Reachy Mini users! Now, you can run the entire conversation app locally, eliminating the need for server audio processing. This new setup utilizes a speech-to-speech pipeline with a Realtime API-compatible WebSocket. Recommended components include llama.cpp with Gemma 4, Silero VAD, Parakeet-TDT STT, and Qwen3-TTS. #ReachyMini #TechUpdate #SpeechTechnology #OpenSource #Innovation
Source: Hugging Face Blog
2026-05-27 00:00
π A recent article discusses the innovative approach of using a Hub Bucket for shipping a trillion parameters in Delta Weight Sync at TRL. This method enhances efficiency and improves data management in large-scale operations. The implications for technology and logistics are significant as organizations strive for better performance. #DataManagement #Logistics #TechInnovation #DeltaWeightSync #TRL
Source: Hugging Face Blog
2026-05-25 00:00
Navigating the evolving landscape of AI Agents can be challenging due to rapidly changing terminology. π§ A recent article addresses common terms like "harness" and "scaffold," highlighting confusion even among experienced practitioners. It aims to clarify these concepts without being an exhaustive dictionary. π Understanding these terms is crucial for anyone involved in building or using AI tools. #AI #Terminology #KnowledgeSharing #AIAgents #TechTrends
Source: Hugging Face Blog
2026-05-23 00:02
π Exciting advancements in language models are here! The article discusses Nemotron-Labs' approach to enhancing text generation speed through diffusion language models. Traditional large language models generate text autoregressively, creating a limit on processing speed. Nemotron-Labs aims to overcome this limitation, potentially transforming how we interact with AI in coding, summarization, and more. #AI #LanguageModels #TextGeneration #Innovation #TechTrends
Source: Hugging Face Blog
2026-05-22 15:25
In AI procurement, the article highlights that specialization often holds more value than sheer scale. Organizations should consider niche expertise when selecting AI solutions, as specialized providers may deliver better results tailored to specific needs. Prioritizing strategic fit over size can enhance effectiveness in AI implementation. π€π #AI #Procurement #Specialization #BusinessStrategy #TechTrends
Source: Hugging Face Blog
2026-05-19 18:38
π Exciting updates on OlmoEarth v1.1! Since its launch in November 2025, this model has been effectively used for various environmental tasks. It tracks mangrove changes, classifies forest loss drivers, and produces crop-type maps rapidly. The advancements in this version aim to enhance efficiency for organizations dedicated to environmental protection. π Explore the models, tech report, and code here: π§ [Models](https://huggingface.co/collections/allenai/olmoearth) π [Tech...
Source: Hugging Face Blog
2026-05-19 00:00
π Exciting news in AI! The Ettin Reranker Family has been introduced, bringing advancements in information retrieval. These new models aim to improve the ranking of search results for better user experience. Stay tuned for updates on their performance and applications! #AI #MachineLearning #InformationRetrieval #EttinReranker #TechNews
Source: Hugging Face Blog
2026-05-18 16:00
NVIDIA has introduced advancements in video generation for robots by fine-tuning Cosmos Predict 2.5 using LoRA and DoRA techniques. These enhancements aim to improve the quality and efficiency of video outputs, making robotic applications more effective. This development highlights the growing intersection of AI and robotics. π€π₯ #NVIDIA #Robotics #AI #VideoGeneration #TechInnovation
Source: Hugging Face Blog
2026-05-18 15:12
PaddleOCR 3.5 enhances OCR and document parsing capabilities by integrating with Hugging Face Transformers. πβ¨ Users can now utilize PaddleOCR models, including PP-OCRv5 and PaddleOCR-VL 1.5, seamlessly with the Transformers backend. Explore the live demo on Hugging Face Spaces to see it in action! π #PaddleOCR #OCR #DocumentParsing #HuggingFace #AI
Source: Hugging Face Blog
2026-05-18 14:12
π Exciting news in AI evaluation! The Open Agent Leaderboard has been launched to assess the performance of general-purpose AI agents. Unlike traditional evaluations, it focuses on entire systems, including tools, planning, and recovery methods. This new benchmark highlights both the quality and cost of deploying various agents, providing clearer insights for users. The leaderboard is supported by the Exgentic framework, fostering transparency from the start. #AI #OpenBenchmark...
Source: Hugging Face Blog
2026-05-14 18:55
Introducing Granite Embedding Multilingual R2, a new open-source project under Apache 2.0. This initiative focuses on multilingual embeddings with an impressive 32K context. Research highlights that it offers the best retrieval quality for datasets under 100 million. This development aims to enhance multilingual processing in various applications. Stay tuned for further updates on advancements in multilingual technology! ππ #Multilingual #OpenSource #Apache2 #DataScience #TechInnovation
Source: Hugging Face Blog
2026-05-14 00:00
Unlocking asynchronicity in continuous batching can lead to significant performance gains for inference. This article discusses how to separate CPU and GPU workloads to optimize GPU utilization. Continuous batching has improved GPU efficiency, but it remains synchronous, causing idle time. By implementing asynchronous batching, CPU preparation and GPU computation can operate simultaneously, reducing wasted time and enhancing performance. ππ»π #AsynchronousBatching #GPUUtilization...
Source: Hugging Face Blog
2026-05-11 23:18
π The landscape of foundation model training is shifting on AWS. Previously, scaling focused on pre-training with more compute leading to better performance. However, recent insights show that scaling now involves post-training methods like supervised fine-tuning and reinforcement learning. Additionally, optimizing test-time compute strategies is becoming crucial for further enhancements. #AWS #MachineLearning #FoundationModels #AI #Scaling
Source: Hugging Face Blog