DeepSeek has released its V4 family of open-weight large language models, featuring a 1.6 trillion parameter model and a smaller 284 billion parameter Flash MoE model. These new models claim to rival top proprietary LLMs in performance while significantly reducing inference costs. Key to this efficiency are architectural innovations like a hybrid attention mechanism and the use of lower precision datatypes (FP8 and FP4), enabling a million-token context window with substantially less memory. AI
影响 Sets new efficiency benchmarks for open-weight models, potentially lowering inference costs and enabling larger context windows for a wider range of applications.
排序理由 Release of new open-weight LLMs from a notable AI lab with claims of rivaling proprietary models and significant efficiency gains.
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
