The Qwen2 Series, Alibaba Cloud’s most recent language model series, has outperformed every other open-source language learning model (LLM) in various benchmarks to emerge at the top of the LLM Leaderboard. Trained with nearly 30 languages, Alibaba’s LLM surpassed the competition soon after its launch, thanks to its enhanced safety alignment as well as the improved performance it offers.
This latest LLM series from Alibaba Group Holding Ltd. (NYSE: BABA) comprises several instruction-turned language models ranging from 0.5 to 72 billion parameters in size, base languages and a Mixture-of-Experts (MoE). Thanks to these updated capabilities, Qwen2 raced to the number-one spot on Hugging Face’s Open LLM Leaderboard. Hugging Face is a collaborative artificial intelligence platform where Qwen2 is currently available for both research and commercial purposes.
According to Zhou Jingren, chief technology officer at Alibaba Cloud, the company hopes to develop the “most open cloud in the AI era,” make the burgeoning artificial intelligence segment more accessible and increase computing power inclusivity. With technology companies such as OpenAI developing more advanced AI by the day, Alibaba Cloud and other cloud providers are already working to integrate the technology into their systems.
Alibaba Cloud plans to pull ahead of the pack with offerings such as the Qwen2, which is available on the collaborative AI platform Hugging Face as well as ModelScope, Alibaba’s AI model community. Alibaba leveraged optimized training methods in the development of the Qwen2-72B model, allowing it to beat other top open-source language learning models in 15 benchmarks, including reasoning, mathematics, multilingual capability, language generation and coding.
The model has also displayed the capacity to handle up to 128K tokens, the highest number of tokens a language model can remember while it generates text. Alibaba Cloud used 27 languages including English, Chinese, Spanish, Italian, Hebrew, Persian, German and Arabic to boost Qwen2’s multilingual abilities. The Chinese technology company also used a technique called Group Query Attention to increase the Qwen2 Model’s speed and lower memory use by optimizing the balance between model performance and computational efficiency.
Aside from Qwen2’s extremely high-level linguistics and mathematics capabilities, its output also indicates that the model aligns better with human values, giving it one more edge over other open-source language learning models. The MT-bench revealed that Qwen2 scored highly in instruction following and multiturn conversational ability, two elements that are crucial to a chatbot’s interactions with people.
Alibaba included human feedback to help the Qwen2 model align with human values better and perform better in terms of responsibility and safety. This allows Alibaba’s LLM to deal with unsafe multilingual queries associated with criminal activities, such as privacy violations and fraud, and prevents it from being misused by bad players.
About ChineseWire
ChineseWire (“CW”) is a specialized communications platform with a focus on promising China-based companies that are listed in North America. It is one of 60+ brands within the Dynamic Brand Portfolio @ IBN that delivers: (1) access to a vast network of wire solutions via InvestorWire to efficiently and effectively reach a myriad of target markets, demographics and diverse industries; (2) article and editorial syndication to 5,000+ outlets; (3) enhanced press release enhancement to ensure maximum impact; (4) social media distribution via IBN to millions of social media followers; and (5) a full array of tailored corporate communications solutions. With broad reach and a seasoned team of contributing journalists and writers, CW is uniquely positioned to best serve private and public companies that want to reach a wide audience of investors, influencers, consumers, journalists and the general public. By cutting through the overload of information in today’s market, CW brings its clients unparalleled recognition and brand awareness. CW is where breaking news, insightful content and actionable information converge.
For more information, please visit https://www.ChineseWire.com
Please see full terms of use and disclaimers on the ChineseWire website applicable to all content provided by CW, wherever published or re-published: https://www.ChineseWire.com/Disclaimer
ChineseWire
Los Angeles, CA
www.ChineseWire.com
310.299.1717 Office
[email protected]
ChineseWire is powered by IBN
