Open Source News & Articles on Techminis

A naukri.com initiative

New

Home

Open Source News

Marktechpost

132

IBM and Hugging Face Researchers Release SmolDocling: A 256M Open-Source Vision Language Model for Complete Document OCR

IBM and Hugging Face Researchers have released SmolDocling, a 256M open-source vision language model (VLM) for document OCR.
SmolDocling provides a streamlined solution for end-to-end multi-modal document conversion tasks, processing entire pages through a single model.
It utilizes a universal markup format called DocTags to capture page elements and structures, and achieves high performance in benchmark tests.
SmolDocling is capable of handling diverse elements within documents and offers comprehensive structured metadata for enhanced usability.

Read Full Article

7 Likes

VentureBeat

269

Image Credit: VentureBeat

Nvidia’s GTC 2025 keynote: 40x AI performance leap, open-source ‘Dynamo’, and a walking Star Wars-inspired ‘Blue’ robot

Nvidia's GTC 2025 keynote was delivered by CEO Jensen Huang at the SAP Center and showcased the company's future vision in AI and robotics.
The event highlighted the Blackwell platform's full production, offering a 40x AI performance leap over its predecessor, Hopper.
Nvidia addressed the demand for efficient AI reasoning models like DeepSeek's R1 by emphasizing increased computation requirements and new AI infrastructures.
A detailed roadmap up to 2027 was presented, outlining future AI computing infrastructure products like Blackwell Ultra, Vera Rubin, and Rubin Ultra.
Nvidia also introduced Dynamo, an open-source system to optimize AI inference, positioning it as a fundamental technology for the AI revolution.
The company unveiled 'Blue,' a Star Wars-inspired robot, signaling its foray into robotics and physical AI to address labor shortages and market opportunities.
Partnerships with Google DeepMind, Disney Research, and General Motors were announced to advance open-source models and technologies in robotics and autonomous vehicles.
Nvidia's strategy expands its AI reach from data centers to manufacturing, enterprise, and self-driving cars, showcasing a comprehensive approach to AI implementation.
Huang emphasized Nvidia's position as an end-to-end AI infrastructure company, showcasing a vision beyond hardware with a focus on software optimization and simulation for AI development.
The event illustrated Nvidia's commitment to driving the AI revolution forward, transitioning computing paradigms from servers to physical devices in industries like automotive and robotics.
Despite some investor skepticism following the event, Nvidia's strategic moves in AI, robotics, and autonomous vehicles demonstrate its dedication to shaping the future of technology.

Read Full Article

16 Likes

VentureBeat

Image Credit: VentureBeat

Nvidia debuts Llama Nemotron open reasoning models in a bid to advance agentic AI

Nvidia has introduced a new set of open source Llama Nemotron reasoning models to accelerate agentic AI workloads.
The reasoning models are based on Nvidia's Nemotron models and are competitive with DeepSeek models, offering business-ready AI reasoning models for advanced agents.
Nvidia's Llama Nemotron models have exceptional reasoning capabilities across math, tool calling, instruction following, and conversational tasks.
Nvidia also announced the Agent AI-Q blueprint, an open-source framework that enables AI agents to query multiple data types and connect to enterprise systems.

Read Full Article

4 Likes

Nvidia

Image Credit: Nvidia

NVIDIA Aerial Expands With New Tools for Building AI-Native Wireless Networks

NVIDIA has expanded its Aerial Research portfolio with new tools for building AI-native wireless networks.
The tools include Aerial Omniverse Digital Twin, Aerial Commercial Test Bed, Sionna 1.0 open-source library, and Sionna Research Kit.
These tools aim to accelerate AI-RAN and 6G research to meet the demands of AI-enabled devices and autonomous vehicles.
Industry leaders and research institutions are already utilizing the NVIDIA Aerial Research portfolio to develop groundbreaking wireless innovations.

Read Full Article

4 Likes

Nvidia

192

Image Credit: Nvidia

NVIDIA Unveils Open Physical AI Dataset to Advance Robotics and Autonomous Vehicle Development

NVIDIA has introduced an open-source dataset aimed at advancing robotics and autonomous vehicle development by providing vast high-quality data.
This dataset includes trajectories for robotics training and Universal Scene Description assets, with future support for autonomous vehicles.
The dataset supports various AI applications such as robots navigating warehouse environments, humanoid robots, and autonomous vehicles.
It aims to become the largest open dataset for physical AI development, benefiting researchers and developers in various domains.
The dataset can enhance AI model performance through pretraining and post-training, offering diverse scenarios and real-world physics representation.
It addresses the challenge of data collection and annotation for AI development, particularly in the field of autonomous vehicles.
Developers can utilize tools like NVIDIA NeMo Curator to efficiently process vast datasets for model training and customization.
University labs, including UCSD and Berkeley DeepDrive, are set to adopt the dataset for research in robotics, autonomous systems, and safety evaluations.
Christensen at UCSD aims to develop semantic AI models for robots in various environments, while CMU's Safe AI Lab plans to evaluate self-driving car safety.
Berkeley DeepDrive and CMU researchers see the dataset as valuable for training AI models with causal reasoning and addressing edge cases in autonomous systems.

Read Full Article

11 Likes

Nvidia

227

Image Credit: Nvidia

NVIDIA Open-Sources cuOpt, Ushering in New Era of Decision Optimization

NVIDIA has open-sourced cuOpt, an AI-powered decision optimization engine, allowing developers to unlock real-time optimization at a large scale across industries like logistics, retail, and airlines.
Key players in the optimization ecosystem such as COPT, FICO's Xpress team, and IBM are integrating or evaluating cuOpt, enhancing decision-making speed.
NVIDIA collaborates with COIN-OR Foundation, enabling cuOpt to become open source and accessible to a wider community for solving complex optimization problems.
cuOpt enables real-time decision-making for airlines, power grids, and financial institutions, optimizing operations efficiently.
By leveraging NVIDIA GPUs, cuOpt evaluates billions of variables simultaneously, significantly accelerating computations and finding optimal solutions faster.
Traditional optimization approaches navigate solution spaces sequentially, but cuOpt evaluates millions of possibilities intelligently, exponentially enhancing optimization speed.
cuOpt accelerates linear programming and mixed-integer programming tasks significantly, showcasing speedups up to 3,000x in various optimization scenarios.
Enhanced optimization not only improves business efficiency but also contributes to a more sustainable, resilient, and equitable world by reducing waste and enabling real-time resource allocation.
Industry leaders like FICO, Gurobi Optimization, and IBM are exploring GPU-accelerated optimization with cuOpt, ushering in a new era for decision intelligence.
cuOpt revolutionizes decision-making processes, providing developers with a high-performance AI toolkit and researchers with new opportunities for pushing AI-driven decision-making boundaries.
cuOpt, to be released as open source later this year, offers real-time optimization benefits by integrating with NVIDIA AI Enterprise software platform for easy deployment in diverse environments.

Read Full Article

13 Likes

Unite

180

Image Credit: Unite

Open-Source Alternatives Amid Semgrep Licensing Controversy

Rival companies launch Opengrep, a fork of Semgrep, an open-source static application security testing tool.
Opengrep aims to provide unrestricted commercial and public access to its code, as an alternative to Semgrep's restrictive licensing model.
DevSecOps startup DeepSource launches Globstar, an open-source toolkit for code security, backed by Y-Combinator investors.
Other open-source alternatives for code analysis include SonarQube and ShellCheck, providing developers and enterprises with different options.

Read Full Article

10 Likes

Hackernoon

270

Image Credit: Hackernoon

The HackerNoon Newsletter: Can ChatGPT Predict the Future? (3/18/2025)

Open Source Is Supposed to Be a Meritocracy—So Why Are Companies Trying to Buy Their Way In?
TechWomen Fellows Fight to Save U.S.-Backed STEM Exchange Program
Can ChatGPT Predict the Future?
AI Honeypots Are the Future of Cybersecurity

Read Full Article

16 Likes

Siliconangle

Image Credit: Siliconangle

What to expect at Chainguard Assemble: Join theCUBE Mar. 26

As software development becomes more complex, enterprises face challenges with software supply chain security due to vulnerabilities introduced by open-source components.
Chainguard Inc. is leading the charge in promoting a secure-by-default model to address software security issues, focusing on reducing vulnerabilities proactively.
The Chainguard Assemble event will showcase the company's commitment to proactive security and feature discussions on software security innovations.
The event is expected to provide insights into the evolving software security landscape, with major announcements from Chainguard.
The Coalition for Secure AI (CoSAI), spearheaded by industry leaders like Chainguard, aims to develop tools for securing AI applications amidst rising security threats.
Chainguard's $140 million Series C funding supports the launch of Chainguard AI Images, enhancing security measures for AI applications.
The company offers hardened, containerized versions of open-source tools to mitigate vulnerabilities at their source and challenge traditional security models.
As enterprises integrate AI into their workflows, the need for secure AI systems grows, with Chainguard contributing to setting new benchmarks for AI security.
Proactive, built-in security measures are gaining traction as traditional security methods struggle to keep pace with advancing threats, shaping the future of secure software development.
TheCUBE will provide exclusive coverage of the Chainguard Assemble event on March 26, featuring discussions with industry experts on software security and vulnerability prevention.

Read Full Article

4 Likes

Hackernoon

223

Image Credit: Hackernoon

Open Source Is Supposed to Be a Meritocracy—So Why Are Companies Trying to Buy Their Way In?

Open source projects frequently attribute contributions to individual identities rather than company affiliations. Individuals are driven by intrinsic factors such as passion for development and extrinsic factors like career advancement, while companies are motivated by goals like quality improvement and innovation in contributing to open source projects.
Open source projects operate as meritocracies, where influence is earned through contributions, in contrast to the hierarchical decision-making structures commonly found in companies.
Establishing a strong organizational identity through consistent contributions in open source projects helps companies build trust and influence over time within the community.
Challenges like burnout among maintainers in open source projects highlight the need for better governance structures and workload distribution to ensure project continuity and individual well-being.
Organizational identity theory can be applied to open source, showing that merit in open source is commonly linked to individual identities due to their more constant nature compared to fluid organizational identities.
Corporations are increasingly contributing to open source projects due to motivations such as improving software quality, fostering innovation, and gaining strategic influence in the long term.
The tension between individual and organizational merit in open source projects underscores the importance of having a well-defined open source strategy for companies to effectively engage in the community.
Individual contributions in open source are driven by a mix of intrinsic and extrinsic motivations, while company contributions focus on enhancing software quality, fostering innovation, and securing long-term influence.
The relationship between individual and company identities in open source projects plays a crucial role in shaping the dynamics of contributions and governance within these communities.
Balancing the principles of meritocracy with hierarchical decision-making structures and addressing challenges like burnout among maintainers are essential for the sustainable development and success of open source projects.

Read Full Article

13 Likes

Marktechpost

412

ByteDance Research Releases DAPO: A Fully Open-Sourced LLM Reinforcement Learning System at Scale

ByteDance, Tsinghua University, and the University of Hong Kong have released DAPO, an open-source reinforcement learning system called Dynamic Sampling Policy Optimization for Large Language Models (LLMs).
DAPO aims to enhance the reasoning abilities of LLMs and promote reproducibility by openly sharing algorithmic details, training procedures, and datasets.
DAPO incorporates four core innovations: Clip-Higher, Dynamic Sampling, Token-level Policy Gradient Loss, and Overlong Reward Shaping.
Experimental results demonstrate significant improvements with DAPO, achieving higher scores on the American Invitational Mathematics Examination (AIME) 2024 benchmark with half the training steps compared to previous methods.

Read Full Article

24 Likes

Medium

232

Image Credit: Medium

Mistral Introduces Mistral Small 3.1: The Best Model in Its Weight Class

Mistral Small 3.1 by Mistral AI is a groundbreaking AI model that offers unmatched performance and features.
It is multimodal, multilingual, and fast, with a 128k token context window, processing 150 tokens per second.
Outperforming competitors like Gemma 3 and GPT-4o Mini, Mistral Small 3.1 runs efficiently on minimal hardware.
Its open-source nature under the Apache 2.0 license allows free usage for personal and commercial projects.
The model excels in handling multimodal inputs, supporting multiple languages, making it versatile for various applications.
It outshines Gemma 3 and GPT-4o Mini in accuracy, efficiency, and latency, boasting high performance with fewer parameters.
Mistral Small 3.1 can process 150 tokens per second, ideal for real-time applications like virtual assistants.
With optimized hardware requirements, it can run on a single RTX 4090 GPU or a Mac with 32GB RAM locally.
The model is available on Hugging Face, Mistral AI's developer playground, Google Cloud, and soon on NVIDIA NIM.
Its open-source Apache 2.0 license fosters collaboration and innovation, positioning Mistral Small 3.1 as a game-changer in the field of AI.

Read Full Article

13 Likes

VentureBeat

361

Image Credit: VentureBeat

Mistral AI drops new open-source model that outperforms GPT-4o Mini with fraction of parameters

French startup Mistral AI has introduced a new open-source model, Mistral Small 3.1, boasting better performance with just 24 billion parameters, in contrast to leading proprietary models.
The model offers improved text and image processing, multimodal understanding, and an expanded context window of 128k tokens, achieving speeds of 150 tokens per second.
Mistral’s approach of releasing models under the Apache 2.0 license contrasts with competitors' restricted access, emphasizing open accessibility in the AI industry.
Founded by former Google DeepMind and Meta researchers, Mistral AI has gained a $6 billion valuation, emphasizing European digital sovereignty and sustainability in AI development.
Mistral’s efficiency in using smaller architectures addresses computational and energy cost concerns associated with large models, making AI more accessible for on-device applications.
The company’s strategic focus on efficiency and optimization rather than sheer scale aligns with sustainability goals in the face of evolving industry challenges.
As European AI startups like Mistral gain momentum, concerns about global AI competition are shifting, with a potential for European alternatives to gain market traction.
Emphasizing European digital sovereignty, Mistral stands out for its regulatory alignment with EU values, contrasting with the challenges faced by American and Chinese competitors.
Mistral’s diverse AI product portfolio, including specialized tools like Mistral OCR and Saba, demonstrates a strategic balance between innovation and market demands.
Bridging partnerships and alliances, including deals with Microsoft and strategic players in Europe, positions Mistral as a key player in the region's AI ecosystem.
While Mistral's open-source strategy challenges traditional AI development norms, the company faces revenue challenges and the need to differentiate in a competitive industry.

Read Full Article

21 Likes

The New Stack

391

Image Credit: The New Stack

Meet Kagent, Open Source Framework for AI Agents in Kubernetes

Solo.io introduces kagent, an open source framework designed for building and running AI agents in Kubernetes to streamline workflows.
Kagent caters to DevOps and platform engineers by offering tools, resources, and AI agents for automating tasks like configuration, troubleshooting, observability, and network security.
It integrates with other cloud native tools through the Model Context Protocol (MCP), aiming to standardize AI model integration with APIs.
Based on Microsoft's AutoGen framework, kagent is open source with an Apache 2.0 license.
Initially an internal solution, kagent arose from a customer issue where Solo.io's expertise in Istio and Envoy was leveraged to resolve network problems for an insurance company.
Solo.io plans to contribute kagent to the Cloud Native Computing Foundation (CNCF) after donating Gloo Gateway in November.
Kagent's launch includes tools for Argo, Helm, Istio, Kubernetes, Grafana, Prometheus, and a cloud native expert knowledge base.
The framework comprises three layers: tools, agents for autonomous tasks like canary deployments and security policies, and a declarative API for building and running agents.
Solo.io envisions kagent as a community-driven project with scalability through additional agents contributed by the ecosystem.
Future plans for kagent include tracing capabilities, metrics expansion, multi-agent support, and broadening support for large language models.

Read Full Article

23 Likes

Hackernoon

228

Image Credit: Hackernoon

AI’s Data Bottleneck Is Getting Worse—Here’s a New Way to Fix It

VTS (Vector Transport Service) is an open-source tool developed by Zilliz for vector and unstructured data migration based on Apache SeaTunnel.
Vector databases are designed for high-dimensional vector data storage, used in AI applications like image retrieval, recommendation systems, and natural language processing.
Challenges like data fragmentation and diverse formats exist in dealing with unstructured data in AI applications.
Vendor lock-in in the vector database field can limit organizations' flexibility and innovation, affecting performance and scalability.
Zilliz addresses challenges in vector data migration with its open-source migration tool, VTS, supporting real-time and batch data migration.
VTS supports cross-platform data integration, various data types, and provides excellent performance in data migration tasks.
Future plans for VTS include supporting more data sources, direct insertion of raw data, and integrating with other platforms like Milvus and Apache DolphinScheduler.
VTS simplifies unstructured data conversion through AI models, reduces data cleaning costs, and ensures end-to-end data quality.
The tool aims to streamline vector data migration processes and strengthen support for unstructured data types like images, texts, and PDFs.
Overall, VTS emerges as a crucial data migration tool in AI applications, offering powerful data processing and management capabilities.

Read Full Article

13 Likes

For uninterrupted reading, download the app

Open Source News

Discover more