国内AIGC企业上海岩芯数智人工智能科技有限公司（岩芯数智，RockAI）于1月24日下午在上海发布了国内首个自研的非Transformer Attention机制的低算力通用自然语言大模型——Yan模型。这款大模型在记忆能力、速度和推理吞吐量方面均有显著提升，实现了用百亿级参数达成千亿参数大模型的性能效果。
Yan模型采用了非Transformer Attention机制，相较于传统的Transformer Attention大模型，其在性能上有明显优势。 memory capability 提升了3倍，speed 提升了7倍，inference throughput 提升了5倍。这款大模型的发布，标志着我国在通用大模型领域取得了重要突破。
岩芯数智公司表示，Yan模型的研发目标是满足低算力设备对高性能自然语言处理的需求，推动人工智能技术在更广泛的场景中应用。这款大模型有望在自然语言处理、语音识别、机器翻译等领域发挥重要作用，为我国人工智能产业的发展注入新动力。
英文翻译：
News Title: Domestic large-scale model with non-Transformer Attention mechanism released
Keywords: RockAI, non-Transformer Attention, large-scale model, Yan
News Content:
Shanghai RockAI, an domestic AIGC company, released the first domestically developed large-scale model with a non-Transformer Attention mechanism, named Yan, on January 24th. This large-scale model significantly improves memory capability, speed, and inference throughput, achieving performance comparable to that of a 100-billion-parameter large-scale model with only 100-billion-parameter parameters.
The Yan model adopts a non-Transformer Attention mechanism, which has obvious advantages over traditional Transformer Attention large-scale models in terms of performance. The memory capability has been increased by 3 times, the speed has been increased by 7 times, and the inference throughput has been increased by 5 times. The release of the large-scale model marks an important breakthrough in the field of general-purpose large-scale models in China.
RockAI said that the research and development goal of the Yan model is to meet the demand for high-performance natural language processing on low-computing power devices, and to promote the application of artificial intelligence technology in a wider range of scenarios. This large-scale model is expected to play an important role in fields such as natural language processing, speech recognition, and machine translation, and inject new momentum into the development of China’s artificial intelligence industry.
【来源】https://www.tmtpost.com/6898099.html
Views: 2