【券商聚焦】华泰证券指Grok-3已证明预训练依然有突破空间 持续看好全球头部大模型预训练算力需求

金吾财讯
20 Feb

金吾财讯 | 华泰证券表示,2025年2月18日,xAI发布了基于全球最大算力集群(约20万块GPU)训练的新一代大模型:1)预训练模型:Grok-3与Grok-3 mini,在Math、Science、Coding测评集上超过了Gemini-2 Pro、GPT-4o以及DeepSeek-V3;2)推理模型:Grok-3的Reasoning版本在性能上超越了OpenAI o3 mini (high)。Grok-3系列将首先面向X平台的Premium Plus订阅用户开放,包括聊天功能、DeepSearch与高级推理模式。该行认为,一方面Grok-3训练集群已达到20万卡级别,证明预训练在算力提升下仍能突破;另一方面,xAI将采用“延迟开源”策略,即在Grok-3打磨完善后,开源Grok-2,实现了商业化和开源的平衡,开源社区有望被持续赋能。该行提到,xAI指出,Big intelligence requires big compute,Grok-3相比Grok-2提升了10x的训练算力。2024年9月3日,马斯克宣布xAI的10万卡H100集群上线,本次发布会再次表示xAI的训练集群GPU数已扩大到了20万。该行认为,虽然目前在预训练数据上存在瓶颈,但是合成数据、RL数据、工程能力优化,甚至Transformer架构迭代,或能逐步打破瓶颈,Grok-3已经证明预训练依然有突破空间,持续看好全球头部大模型预训练算力需求。

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.

Most Discussed

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10