China's Alibaba unveils 'Qwen2 (Q12)' as an open

China's Alibaba unveils 'Qwen2 (Q12)' as an open source that surpasses 'Rama 3' and supports Hangeul (Source: aitimes/blog, attached to URL comments)

Alibaba has unveiled its new open source large language model (LLM) called "Qwen2." Some of its performances claimed to outperform Meta's "Rama 3."
The second version of 'Tongyi Qianwen' from Q1.2.
Q1 2 features pre-training in 27 languages other than English and Chinese, and a longer-length context window. (Including Korean)
It provides five models: ▲Q12-0.5B ▲Q12-1.5B ▲Q12-7B ▲Q12-57B-A14B ▲Q12-72B. In particular, Q12-0.5B is considered the current minimum parameter model.
The Q12-72B outperformed the Lamar3-70B, and the smaller Q12 models were similarly high, especially in coding and Chinese-related indicators, it said.
With its previous Q1, it gained widespread popularity in the open-source community. Therefore, analysts say that there are high expectations for the sequel as well.
Alibaba launched Q1 2 a month after unveiling Tungy Chen One 2.5.
At the time, the company said that Tung Ying Chen One 2.5 performs better in Chinese-related functions than "GPT-4," the most advanced model of OpenAI. The model has already been adopted by more than 90,000 companies and has been downloaded more than 7 million times.

*License: Qwen2-72B and instruction tuning models still use traditional Qianwen licenses, but all other models, including Qwen2-0.5B, Qwen2-1.5B, Qwen2-7B, and Qwen2-57B-A14B, will adopt Apache 2.0! We believe that increasing model openness to the community can accelerate Qwen2's application and commercial use around the world.