RSS Reader

非營利的AI研究組織EleutherAI上週發布了Common Pile v0.1 ，這是個AI訓練資料集，所蒐集的資料全部屬於開放授權或公有領域，且利用該資料集訓練出的2個7B模型，發現它們的表現並不亞於LLaMA、MPT、PRJ-INCITE或Qwen等知名模型。

EleutherAI創立於

Open Source LLM

Hugging Face開源機器人模型SmolVLA　可跑在Mac電腦上

ITHome | June 9, 2025, 12:04 a.m. | Original

Hugging Face積極拓展機器人業務，繼公佈二款低成本機器人後，上週再開源一款4.5億參數量的機器人模型SmolVLA ，號稱能執行在消費硬體如Mac電腦上。

相較於AI快速進步，機器人進展緩慢許多，這是因為欠缺高品質多元資料，及協助機

Open Source LLM

NumPy 2.3 Introduces OpenMP Parallelization Support

Phoronix | June 7, 2025, 5 p.m. | Original

NumPy 2.3 is out today as the latest release of this widely-used library for scientific computing. Notable with NumPy 2.3 is finally adding some parallelization support via OpenMP.

NumPy 2.3 brings initial support for building the library with OpenMP parallelization. For NumPy 2.3, the OpenMP support is disabled by default but during build time can be easily enabled with "-Denable_openmp=true". With the v2.3 milestone just the np.sort and np.argsort functions are

Open Source LLM

Beijing academy unveils ‘RoboBrain’ AI model for China’s humanoids

SCMP | June 6, 2025, 7 p.m. | Original

Visitors watch a robot performing during the world robot carnival in Wuhan this week. Photo: Xinhua

Published: 7:00am, 7 Jun 2025

The Beijing Academy of Artificial Intelligence (BAAI), a non-profit research laboratory in China, launched on Friday a series of new open-source artificial intelligence (AI) models designed to be the “brain” of robots, as the country rushes to build smarter machines.

The use of powerful AI models in China’s booming robotics industry could accelerate the development and adoption of humanoids, as the sector addresses challenges such as limited model capabilities

Open Source LLM

Alibaba unveils new AI embedding models, a field it leads globally

SCMP | June 6, 2025, 3 a.m. | Original

Published: 3:00pm, 6 Jun 2025

Alibaba Group Holding has made its Qwen3 Embedding series available for developers, in the Chinese tech giant’s latest bid to solidify its global leadership in open-source artificial intelligence (AI) models.

Released late on Thursday, the series marks another addition to the company’s line-up of large language models (LLMs), which are among the world’s most popular open-source AI systems , according to New York-based

Open Source LLM

OpenAI finds more Chinese groups using ChatGPT for malicious purposes - Reuters

Reuters | June 5, 2025, 5:08 a.m. | Original

OpenAI finds more Chinese groups using ChatGPT for malicious purposes Reuters

Open Source LLM

Gemini 2.5 多國語言實時語音生成 Google 示範人聲幾可亂真

UNWIRE.HK | June 4, 2025, 9 a.m. | Original

Google 正式發布 Gemini 2.5 的進階語音對話和生成功能，標誌著 AI 語音技術的重大突破。這項新功能支援超過 24 種語言的即時語音對話，並提供前所未有的語音控制能力，讓開發者能夠創建更豐富的互動應用程式。Gemini 2.5 現已整合

Open Source LLM

DeepSeek 新模型再被指以「蒸餾」訓練　未經許可使用 Google Gemini 數據

UNWIRE.HK | June 4, 2025, 7:20 a.m. | Original

SUQIAN, CHINA - JANUARY 28: In this photo illustration, the logo of Deepseek is displayed on a smartphone screen on January 28, 2025 in Suqian, Jiangsu Province of China. (Photo by VCG/VCG via Getty Images)

DeepSeek 早前低調推出的最新推理模型 R1-0528 在數學和編程基準測試中表現出色，但隨即引發使用 Google Gemini 數據進行

Open Source LLM

OpenAI 預告 GPT-5 效能大提升　目標 7 月推出應對激烈競爭

UNWIRE.HK | June 4, 2025, 1:02 a.m. | Original

AI 業界競爭相當激烈，OpenAI 高層在一個活動上正式確認下一代基礎模型 GPT-5 正在開發中，並坦承希望透過新模型提升競爭力。這是 OpenAI 首次公開承認在編程等領域面臨 Google Gemini 2.5 Pro 和 Anthropic Claude 4 等競爭對手的挑戰，預計 GPT-5 將於今年

Open Source LLM