对Hugging Face基准测试的最新分析显示，中国以24个头名领先，其次是韩国的12个，美国的3个。一家韩国组织包揽了该国全部12个头名。另外，一款名为Rebellions ATOM-Max的新型推理NPU已被开发出来，其语言模型预训练能力可与GPU媲美，在未见过文本上的得分与Nvidia B200 twin几乎相同。 AI
影响 凸显了全球AI发展趋势和AI模型训练专用硬件的进步。
排序理由 该集群包含AI模型的基准测试结果以及关于一款新型NPU的信息，符合研究类别。
在 Mastodon — fosstodon.org 阅读 →
AI 生成摘要 · Google Gemini · 来自 4 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
Measured on 53 official Hugging Face benchmarks: China 24 # 1 spots, Korea 12, USA 3. Korea's 12 come from one organization on 45 entries. A look at ef # datascience # huggingface # benchmark # ai # software # coding # development # engineering # inclusive # community Korea Holds…
VKORE-325M was pretrained from random init on a Rebellions ATOM-Max inference NPU. On unseen text it scores 0.7993 vs 0.7994 for a B200 twin. What it # ai # machinelearning # edgeai # npu # software # coding # development # engineering # inclusive # community From 46M to 325M: An…
How OpenCode's database, TokenScope, and a fresh greeting helped me trace our agent's token usage to MCP tool definitions—and what Code Mode changed. # ai # mcp # llm # opencode # software # coding # development # engineering # inclusive # community Why Saying ‘Hi’ to Our OpenCod…
When you want a short AI-generated video, there are two common starting points: a ready-made effect... # ai # video # webdev # software # coding # development # engineering # inclusive # community Picking an AI video effect template vs writing a text-to-video prompt