Close Menu
    DevStackTipsDevStackTips
    • Home
    • News & Updates
      1. Tech & Work
      2. View All

      How To Prevent WordPress SQL Injection Attacks

      June 11, 2025

      Creating The “Moving Highlight” Navigation Bar With JavaScript And CSS

      June 11, 2025

      Databricks adds new tools like Lakebase, Lakeflow Designer, and Agent Bricks to better support building AI apps and agents in the enterprise

      June 11, 2025

      Zencoder launches end-to-end UI testing agent

      June 11, 2025

      OpenAI CEO Sam Altman claims “ChatGPT is already more powerful than any human who has ever lived”

      June 11, 2025

      Apple Intelligence delay: A clash of two architectures and trivial AI features fell short of standards and expectations

      June 11, 2025

      Ambrosia Sky is a gorgeous science-fiction game that’s all about death, and I can’t wait to play more

      June 11, 2025

      3 secrets of PowerToys on Windows 11 that you’ll wish you already knew

      June 11, 2025
    • Development
      1. Algorithms & Data Structures
      2. Artificial Intelligence
      3. Back-End Development
      4. Databases
      5. Front-End Development
      6. Libraries & Frameworks
      7. Machine Learning
      8. Security
      9. Software Engineering
      10. Tools & IDEs
      11. Web Design
      12. Web Development
      13. Web Security
      14. Programming Languages
        • PHP
        • JavaScript
      Featured

      [EcjoJS Meta] Content discussion

      June 11, 2025
      Recent

      [EcjoJS Meta] Content discussion

      June 11, 2025

      Accessibility, Inclusive Design, and Universal Design Work Together

      June 11, 2025

      An “Inconceivable” Conversation With Dr. Pete Cornwell on Simple vs. Agentic AI

      June 11, 2025
    • Operating Systems
      1. Windows
      2. Linux
      3. macOS
      Featured

      OpenAI CEO Sam Altman claims “ChatGPT is already more powerful than any human who has ever lived”

      June 11, 2025
      Recent

      OpenAI CEO Sam Altman claims “ChatGPT is already more powerful than any human who has ever lived”

      June 11, 2025

      Apple Intelligence delay: A clash of two architectures and trivial AI features fell short of standards and expectations

      June 11, 2025

      Ambrosia Sky is a gorgeous science-fiction game that’s all about death, and I can’t wait to play more

      June 11, 2025
    • Learning Resources
      • Books
      • Cheatsheets
      • Tutorials & Guides
    Home»Development»Machine Learning»Skywork AI Advances Multimodal Reasoning: Introducing Skywork R1V2 with Hybrid Reinforcement Learning

    Skywork AI Advances Multimodal Reasoning: Introducing Skywork R1V2 with Hybrid Reinforcement Learning

    April 25, 2025

    Recent advancements in multimodal AI have highlighted a persistent challenge: achieving strong specialized reasoning capabilities while preserving generalization across diverse tasks. “Slow-thinking” models such as OpenAI-o1 and Gemini-Thinking have made strides in deliberate analytical reasoning but often exhibit compromised performance on general visual understanding tasks, with increased tendencies toward visual hallucinations. As the field progresses toward building general-purpose AI systems, reconciling this tradeoff remains a critical research problem.

    Skywork AI Introduces Skywork R1V2

    Skywork AI has released Skywork R1V2, a next-generation multimodal reasoning model designed to address the reasoning-generalization tradeoff systematically. Building upon the foundation of Skywork R1V, R1V2 introduces a hybrid reinforcement learning framework, combining reward-model guidance with structured rule-based signals. The model bypasses the conventional reliance on teacher-student distillation by learning directly from multimodal interactions, offering an open and reproducible advancement through its release on Hugging Face.

    Technical Approach and Innovations

    Skywork R1V2 incorporates Group Relative Policy Optimization (GRPO) alongside a Selective Sample Buffer (SSB) to enhance training stability and efficiency. GRPO enables relative evaluation among candidate responses within the same query group, but convergence issues can diminish effective learning signals. The SSB mechanism addresses this by maintaining a cache of informative samples, ensuring continuous access to high-value gradients.

    Additionally, the model adopts a Mixed Preference Optimization (MPO) strategy, integrating reward-model-based preferences with rule-based constraints. This hybrid optimization allows Skywork R1V2 to strengthen step-by-step reasoning quality while maintaining consistency in general perception tasks. A modular training approach, utilizing lightweight adapters between a frozen Intern ViT-6B vision encoder and a pretrained language model, preserves the language model’s reasoning capabilities while optimizing cross-modal alignment efficiently.

    Empirical Results and Analysis

    Skywork R1V2 demonstrates robust performance across a range of reasoning and multimodal benchmarks. On text reasoning tasks, the model achieves 78.9% on AIME2024, 63.6% on LiveCodeBench, 73.2% on LiveBench, 82.9% on IFEVAL, and 66.3% on BFCL. These results represent significant improvements over Skywork R1V1 and are competitive with substantially larger models, such as Deepseek R1 (671B parameters).

    In multimodal evaluation, R1V2 achieves 73.6% on MMMU, 74.0% on MathVista, 62.6% on OlympiadBench, 49.0% on MathVision, and 52.0% on MMMU-Pro. The model consistently outperforms open-source baselines of comparable or larger size, including Qwen2.5-VL-72B and QvQ-Preview-72B, particularly excelling in tasks that require structured problem-solving across visual and textual inputs.

    When compared against proprietary models, R1V2 demonstrates narrowing performance gaps. It surpasses Claude 3.5 Sonnet and Gemini 2 Flash on critical multimodal benchmarks such as MMMU and MathVista. Importantly, hallucination rates were substantially reduced to 8.7% through calibrated reinforcement strategies, maintaining factual integrity alongside complex reasoning.

    Qualitative assessments further illustrate R1V2’s systematic problem-solving approach, with the model demonstrating methodical decomposition and verification behaviors in complex scientific and mathematical tasks, reinforcing its alignment with reflective cognitive patterns.

    Conclusion

    Skywork R1V2 advances the state of multimodal reasoning through a carefully designed hybrid reinforcement learning framework. By addressing the vanishing advantages problem with the Selective Sample Buffer and balancing optimization signals through Mixed Preference Optimization, the model achieves notable improvements in both specialized reasoning tasks and general multimodal understanding.

    With benchmark-leading performances such as 62.6% on OlympiadBench and 73.6% on MMMU, Skywork R1V2 establishes a strong open-source baseline. Its design principles and training methodology offer a pragmatic approach toward developing robust, efficient multimodal AI systems. Future directions for Skywork AI include enhancing general visual understanding capabilities while preserving the sophisticated reasoning foundations laid by R1V2.


    Check out the Paper and Model on HuggingFace. Also, don’t forget to follow us on Twitter and join our Telegram Channel and LinkedIn Group. Don’t Forget to join our 90k+ ML SubReddit.

    🔥 [Register Now] miniCON Virtual Conference on AGENTIC AI: FREE REGISTRATION + Certificate of Attendance + 4 Hour Short Event (May 21, 9 am- 1 pm PST) + Hands on Workshop

    The post Skywork AI Advances Multimodal Reasoning: Introducing Skywork R1V2 with Hybrid Reinforcement Learning appeared first on MarkTechPost.

    Source: Read More 

    Facebook Twitter Reddit Email Copy Link
    Previous ArticleMore Ivanti attacks may be on horizon, say experts who are seeing 9x surge in endpoint scans
    Next Article From GenAI Demos to Production: Why Structured Workflows Are Essential

    Related Posts

    Machine Learning

    How to Evaluate Jailbreak Methods: A Case Study with the StrongREJECT Benchmark

    June 11, 2025
    Machine Learning

    How Gardenia Technologies helps customers create ESG disclosure reports 75% faster using agentic generative AI on Amazon Bedrock

    June 11, 2025
    Leave A Reply Cancel Reply

    For security, use of Google's reCAPTCHA service is required which is subject to the Google Privacy Policy and Terms of Use.

    Continue Reading

    CVE-2025-43570 – Substance3D Use After Free Vulnerability

    Common Vulnerabilities and Exposures (CVEs)

    SpotBugs Access Token Theft Identified as Root Cause of GitHub Supply Chain Attack

    Development

    Hackers Attacking Network Edge Devices to Compromise SMB Organizations

    Security

    The best push mowers of 2025: Expert picks from Cub Cadet, Makita, and more

    News & Updates

    Highlights

    CVE-2025-5379 – NuCom NC-WR744G Console Application Hard-Coded Credentials Remote Vulnerability

    May 31, 2025

    CVE ID : CVE-2025-5379

    Published : May 31, 2025, 2:15 p.m. | 3 hours, 29 minutes ago

    Description : A vulnerability classified as critical was found in NuCom NC-WR744G 8.5.5 Build 20200530.307. This vulnerability affects unknown code of the component Console Application. The manipulation of the argument CMCCAdmin/useradmin/CUAdmin leads to hard-coded credentials. The attack can be initiated remotely. The vendor was contacted early about this disclosure but did not respond in any way.

    Severity: 4.3 | MEDIUM

    Visit the link for more details, such as CVSS details, affected products, timeline, and more…

    Android Show: Google to Unveil OS Future Before I/O 2025

    April 29, 2025

    Google joins OpenAI in adopting Anthropic’s protocol for connecting AI agents – why it matters

    April 10, 2025

    CVE-2025-4802 – GNU C Library LD_LIBRARY_PATH Path Traversal Vulnerability

    May 16, 2025
    © DevStackTips 2025. All rights reserved.
    • Contact
    • Privacy Policy

    Type above and press Enter to search. Press Esc to cancel.