Alibaba Cloud open-sources more LLMs with diverse sizes and multimodal features.

Alibaba Cloud open-sources more LLMs with diverse sizes and multimodal features. (Source – Alibaba Cloud).

Alibaba Cloud boosts open-source community with enhanced AI and more open-sourced LLMs

  • Alibaba Cloud leads AI by supporting Meta’s Llama and open-sourcing advanced LLMs like Qwen-72B and Qwen-1.8B.
  • Alibaba Cloud enhances AI with open-sourced LLMs and new multimodal models.
  • Alibaba Cloud’s launch of powerful LLMs and multimodal models significantly advances AI technology and accessibility.

Alibaba Cloud has recently marked a significant achievement in the AI industry by becoming the first Chinese company to support Meta’s open-source AI model, Llama. This integration allows businesses using Alibaba Cloud in China to harness the capabilities of the Llama model for program development.

Expanding the LLM frontier: Alibaba Cloud’s new releases

Further expanding its footprint in the large language model (LLM) domain, Alibaba Cloud has announced the open sourcing of its own LLMs, Qwen-72B and Qwen-1.8B. Derived from its foundational model, Tongyi Qianwen, these versions feature 72 billion and 1.8 billion parameters, respectively. Now available on platforms like ModelScope and Hugging Face, these releases significantly widen the accessibility to advanced AI technologies.

The company has further expanded its range of AI offerings by introducing additional multimodal LLMs, such as Qwen-Audio and Qwen-Audio-Chat. These are pre-trained and conversationally fine-tuned models focused on audio understanding, catering to research and commercial applications.

Alibaba Cloud’s contributions to AI are extensive, ranging from LLMs of various sizes – 1.8B to 72B parameters – to multimodal models equipped with audio and visual comprehension. Jingren Zhou, CTO of Alibaba Cloud, emphasizes the significance of an open-source ecosystem for the growth of LLMs and AI applications. “We aspire to become the most open cloud and make generative AI capabilities accessible to everyone. To achieve that goal, we’ll continue to share our cutting-edge technology and facilitate the development of the open-source community together with our partners,” Zhou said.

The 72-billion-parameter model, which has been pre-trained on more than 3 trillion tokens, is noted for outperforming other major open-source models across ten benchmarks. These include the Massive Multi-task Language Understanding (MMLU) benchmark, HumanEval for code generation, and GSM8K for arithmetic problems. The model is also proficient in complex tasks like role-playing and language style transfer, enhancing its suitability for various AI applications, such as personalized chatbots.

05122023_Qwen-72B outperforms other major open-source models in ten benchmarks. (Source - Alibaba Cloud).

05122023_Qwen-72B outperforms other major open-source models in ten benchmarks. (Source – Alibaba Cloud).

The Qwen-72B model’s resources, including its code, model weights, and documentation, are freely available for research. Companies with less than 100 million monthly active users can also use the models without charge for commercial purposes.

Alibaba Cloud has made its 1.8-billion-parameter LLM available for edge computing, tailored for devices with limited computational capacity like smartphones. This more compact and efficient model, aimed at providing a cost-effective solution, is currently offered only for research purposes.

Diversifying AI: Multimodal LLMs and audio capabilities

In its commitment to offer LLMs capable of processing diverse input formats, Alibaba Cloud has also released Qwen-Audio and Qwen-Audio-Chat. These models boast enhanced audio understanding capabilities and are designed for research and commercial use. Qwen-Audio can interpret a variety of audio inputs, including human speech and natural sounds, and perform over 30 audio processing tasks. Its fine-tuned version, Qwen-Audio-Chat, is adept at conversational applications, including emotion and tone detection in human speech.

This latest initiative is part of Alibaba Cloud’s ongoing efforts to provide the open-source community with multimodal LLMs beyond text understanding. The company recently launched Qwen-VL and Qwen-VL-Chat, models skilled in visual information processing.

The open sourced models, including Qwen-7B, Qwen-14B, and Qwen-VL, have been downloaded over 1.5 million times on Alibaba Cloud’s ModelScope and Hugging Face since August. ModelScope has grown to be the largest AI model community in China, with many active developers and an impressive download count.

Developers are showing interest with the announcement made from Alibaba Cloud.

Developers are showing interest with the announcement made from Alibaba Cloud. (Source – X).

Alibaba International’s AI vision and Aidge’s launch

Leaders in AI, like Alibaba International, acknowledge the potential of these technologies. Jiang Fan, CEO of Alibaba International, remarks on AI’s transformative role: ” The use of AI has revolutionized everything from product development and management to market insights analysis and customer service.” Alibaba International plans to increase its AI investment to support global commerce, aiming to enhance merchant efficiency and customer satisfaction.

The launch of Aidge, Alibaba International’s AI-powered suite, exemplifies this commitment. Aimed at SMEs, Aidge – an acronym for “AI for Digital and Global Entrepreneurship” – utilizes Alibaba’s proprietary LLMs and computer vision technologies. It addresses market entry challenges for SMEs, helping them create localized marketing content, improve customer service, perform analytical tasks, and optimize search results without language barriers.

Aidge promises to alleviate talent shortages and facilitate international growth for businesses of all sizes. It enables even small companies to explore new markets effortlessly with features that support smooth business adaptation.

Alibaba International is rolling out Aidge services to merchants on its e-commerce platforms, with plans to expand its reach through industry partnerships. Global SaaS providers and ISVs are invited to participate in pilot testing, with a regional rollout expected by year’s end.

The global AI market, set to exceed US$300 billion by 2026, underscores businesses’ need to embrace and capitalize on AI. Aidge empowers SMEs worldwide to conduct business in their native languages, leveraging advanced technology to compete in the global market.

The pilot launch of Aidge is just the beginning of Alibaba International’s AI journey. The company aims to collaborate with industry stakeholders to explore and define AI use cases, focusing on simplifying global business operations for SMEs and unlocking new growth opportunities with AI. This initiative reflects Alibaba International’s dedication to leveraging AI technologies to address the unique challenges SMEs face in the global marketplace.

As Alibaba International continues to pioneer in the AI space, its commitment to AI-driven solutions like Aidge represents a significant stride in enhancing the operational efficiency of SMEs. This approach helps small and medium-sized enterprises overcome language and cultural barriers and positions them competitively in the global market.

The global AI market’s rapid growth highlights the importance of such technological advancements for businesses looking to expand internationally. Alibaba International’s efforts to provide SMEs access to AI tools and technologies are crucial in driving the development of the global e-commerce industry.

The pilot launch of Aidge by Alibaba International is a strategic move in AI aimed at empowering SMEs with advanced, accessible AI tools.

By making advanced LLMs like Qwen-72B and Qwen-1.8B more available, along with introducing multimodal models, the boundaries of AI are being expanded, paving the way for more innovative and inclusive AI solutions. These developments are crucial in a world increasingly reliant on AI, offering new opportunities for businesses, developers, and researchers to leverage AI in varied and impactful ways.