Qwen Family LLM

A Qwen Family LLM is a China-based LLM developed by Alibaba Cloud.

• Context:
  • It can contain LLMs that span from 1.8 billion to 72 billion parameters.
  • It can (typically) be based on a Qwuen Multilingual LLM Corpus, of 3 trillion tokens of multilingual data, emphasizing Chinese and English languages.
  • It can include Vision-Language Models.
  • It can include Audio-Language Models (Qwen-Audio), reflecting its versatility across different applications【GPT-4†source】.
  • It can include Quantized LLMs.
  • ...
• Example(s):
  • Qwen Family LLM, 2023-12.
  • Qwen Family LLM, 2024-01.
  • Qwen Family LLM, 2024-01.
  • ...
• Counter-Example(s):
  • ERNIE (Baidu): Another Chinese-based LLM developed by Baidu, focusing on knowledge-enhanced pretraining for natural language understanding and generation.
  • WuDao: A large-scale multimodal LLM developed by Beijing Academy of Artificial Intelligence (BAAI), aiming at understanding and generating both text and images.
  • PLATO (Baidu): Developed by Baidu, focused on conversational AI with advancements in dialogue understanding and generation.
• See: Multimodal Model, AGI-Related Project.

---

---

References

• https://github.com/QwenLM/Qwen

2024

• https://huggingface.co/Qwen
  • QUOTE: This is the organization of Qwen (abbr. for Tongyi Qianwen 通义千问), which refers to the large language model family built by Alibaba Cloud. In this organization, we continuously release large language models (LLM), large multimodal models (LMM), and other AGI-related projects. Now, we have released Qwen, a large language model series spanning from 1.8 billion parameters to 72 billion parameters. Also, we have released Qwen-VL and Qwen-Audio, which are vision-language model and audio-language model based on Qwen.

GPT-4

• NOTES: ...
  • A Qwen LLM Family is developed by Alibaba Cloud, aiming to advance artificial general intelligence (AGI) through the release of large language models (LLMs) and multimodal models​.

​** A Qwen LLM Family encompasses models ranging from 1.8 billion to 72 billion parameters, including specialized versions for vision-language (Qwen-VL) and audio-language (Qwen-Audio) tasks​ ​** A Qwen LLM Family model is pretrained on up to 3 trillion tokens of multilingual data, covering a broad spectrum of domains and languages, with a particular emphasis on Chinese and English​ ​** A Qwen LLM Family demonstrates competitive performance on a variety of benchmark datasets, excelling in tasks such as chatting, content creation, summarization, translation, and coding​ ​** A Qwen LLM Family includes features like system prompt enhancement and supports roles such as agents or code interpreters, showcasing its versatility across different applications​ ​** A Qwen LLM Family recently introduced Qwen1.5, offering base and chat models across six sizes and including quantized models for efficient performance​ ​** A Qwen LLM Family has made strides in multimodal AI with the Qwen-VL series, enhancing image-related reasoning capabilities and supporting high-definition images of various aspect ratios​ ​** A Qwen LLM Family models are accessible for a wide range of AI applications, from simple inference to complex tasks involving tool use, reflecting their broad utility and adaptability​ ​** A Qwen LLM Family's continuous updates and expansions, such as the introduction of quantized models and enhancements in multimodal capabilities, highlight the team's commitment to pushing the boundaries of current AI technologies​ ​

---