Foresight News 消息，据金十报道，DeepSeek 今日于 AI 开源社区 Hugging Face 上发布了一个名为 DeepSeek-Prover-V2-671B 的新模型。DeepSeek-Prover-V2-671B 使用了更高效的 safetensors 文件格式，并支持多种计算精度，方便模型更快、更省资源地训练和部署，参数达 6710 亿，或为去年发布的 Prover-V1.5 数学模型升级版本。在模型架构上，该模型使用了 DeepSeek-V3 架构，采用 MoE（混合专家）模式，具有 61 层 Transformer 层，7168 维隐藏层。同时支持超长上下文，最大位置嵌入达 16.38 万，使其能处理复杂的数学证明，并且采用了 FP8 量化，可通过量化技术减小模型大小，提高推理效率。
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.