A research team at Alibaba Cloud, known as Qwen, has introduced the LLM model under the name Qwen2, which comes in 5 sizes ranging from 0.5B, 1.5B, 7B, 14B, and 72B. One of its key features is its ability to support languages other than English, such as Thai, Vietnamese, Indonesian, Myanmar, Laos, and Cambodia in the Southeast Asian region, in addition to supporting a context window of up to 128K.
Popular evaluation results like MMLU or HumanEval show that Qwen7-72B outperforms Llama3-70B slightly, while Qwen2-7B has outperformed Llama3-7B in multiple test sets, notably in the HumanEval test where it scored significantly higher.
Qwen2 is available for use under the Apache 2.0 license, except for Qwen2-72B, which is restricted to the Qianwen License. This allows the 7B model to be utilized with almost no limitations.
Both the 7B and 72B versions of Qwen2 can be tested on HuggingFace.
Source: QwenLM
TLDR: Alibaba Cloud’s Qwen research team unveils the versatile Qwen2 LLM model with various sizes and language support, showcasing superior performance in evaluations and offering flexibility for usage under different licenses.
Leave a Comment