# Пов'язані статті щодо Repository

Центр новин HTX надає останні статті та поглиблений аналіз на тему "Repository", що охоплює ринкові тренди, оновлення проєктів, технологічні розробки та регуляторну політику в криптоіндустрії.

First Long-Horizon Doc2Repo Training Dataset: Code Agents Move Beyond Bug Fixing and Begin Creating Repositories

With the advancement of LLM Code Agents, the research focus is shifting towards long-horizon, real-world tasks, moving beyond simple bug fixes to full repository generation. To address this, researchers from Renmin University of China introduced the DeNovoSWE dataset. This dataset focuses on long-term software engineering tasks, specifically the "document-to-repository" challenge—generating an entire, executable code repository from a task description. The DeNovoSWE construction method employs a Divide & Conquer approach. It breaks down target repositories into core capabilities and uses a multi-agent Draft-Critic-Repair workflow to automatically generate high-quality, evaluation-aligned task documents. The dataset also implements difficulty-aware filtering to balance quality and diversity. The result is a high-quality, anti-leakage dataset of 4,818 instances. Experiments show that models trained on DeNovoSWE achieve significant improvements in long-horizon repository generation. For instance, Qwen3-30B-A3B-Instruct's performance on the BeyondSWE-Doc2Repo benchmark increased from 5.8% to 47.2%, and on NL2RepoBench from 4.3% to 23.0%. Similar gains were observed with stronger backbones, demonstrating that dedicated long-horizon training data is crucial for advancing Code Agents from maintainers to architects capable of planning and building complete software projects from scratch.

marsbit06/25 08:51

First Long-Horizon Doc2Repo Training Dataset: Code Agents Move Beyond Bug Fixing and Begin Creating Repositories

marsbit06/25 08:51

GitHub Announces Default Use of Copilot User Data for AI Model Training Starting April 24

GitHub has announced an update to its repository policy, effective April 24, 2026, allowing the use of user interaction data to train its AI models. The data collection will include users of Copilot Free, Pro, and Pro+, covering model inputs and outputs, code snippets, contextual information, repository structures, and chat logs. According to GitHub’s Chief Product Officer Mario Rodriguez, the move aims to enhance the accuracy and security of the model’suggestions, with internal Microsoft tests already showing improved acceptance rates. The policy follows an opt-out model, meaning affected users must manually disable data sharing in their privacy settings, sparking debate within the developer community over data ownership and the definition of private repositories. Copilot Business, Enterprise, and educational users are currently exempt due to contractual terms. GitHub defended the change as consistent with industry practices adopted by companies like Anthropic, JetBrains, and Microsoft. However, the inclusion of private repository code in training sets challenges conventional notions of privacy. This shift reflects a broader industry trend where leading AI providers are turning to user interaction data as high-quality public code resources diminish. It signals GitHub’s continued transition from an open-source platform to a closed-loop AI training ecosystem and highlights growing tensions between data compliance and AI model advancement.

marsbit03/26 01:39

GitHub Announces Default Use of Copilot User Data for AI Model Training Starting April 24

marsbit03/26 01:39

活动图片

Othersright-arrow

Ethereumright-arrow