Alibaba Cloud open-sourced two large language models (LLM), Qwen-72B and Qwen-1.8B. The company also opened availability for more multimodal LLMs, including Qwen-Audio, a pre-trained audio understanding model, and Qwen-Audio-Chat, its version for research and commercial purposes.
Open-source ecosystem
“Building up an open-source ecosystem is critical to promoting the development of LLM and AI applications building. We aspire to become the most open cloud and make generative AI capabilities accessible to everyone,” said Jingren Zhou, CTO of Alibaba Cloud.
Qwen-72B and Qwen-1.8B
Qwen-72B is pre-trained on over 3 trillion tokens and outperforms other major open-source models such as multitask accuracy, code generation capabilities, and arithmetic problem-solving.
Organisations can access the Qwen-72B model’s code, model weights, and documentation for free for research purposes. Companies with fewer than 100 million monthly active users can also access the models for commercial uses.
Alibaba Cloud also open-sourced Qwen-1.8B, the lightweight LLM that can be used on end devices such as cellphones, making it more cost-effective and easy to deploy.
Qwen-Audio and Qwen-Audio-Chat
Alibaba Cloud has also open-sourced Qwen-Audio and Qwen-Audio-Chat, the models with enhanced audio understanding capabilities for research and commercial purposes.
Qwen-Audio understands text and audio input in human speech, natural sound, and music and produces text as output. Qwen-Audio-Chat, its conversationally fine-tuned version, can support multiple rounds of conversations based on audio and can detect emotions and tones in human speeches.