AliBaBa M6

Alibaba M6: Ultra-large-scale Chinese pre-trained model by DAMO Academy

Alibaba M6 Official Website, the ultra-large-scale Chinese pre-trained model launched by Alibaba DAMO Academy

What is Alibaba M6?

M6 is the largest cross-modal pre-trained model in the Chinese community, with over ten trillion parameters, featuring strong multi-modal representation capabilities. M6 integrates information from different modalities into a unified processing, forming knowledge representation, providing intelligent services like language understanding, image processing, and knowledge representation across various industries, promoting the democratization of AI!

Alibaba M6 Official Website:https://m6.aliyun.com/#/

How is Alibaba M6?

Offers best practice cases in big data and AI scenarios, a one-stop guided experience to quickly understand the capabilities and solutions of products like DataWorks, MaxCompute, Machine Learning PAI, Hologres, and Flink, reducing the cost of enterprise cloud migration.
M6 outputs generalization capabilities in the form of pre-trained models, allowing for rapid production of accurate models with industry characteristics through optimization and fine-tuning with scenario-specific data.
Pre-trained large models represent a new model for AI development and usage, lowering the threshold for AI adoption and promoting the application and popularization of AI across various industries.

High Precision, Low Threshold, Multi-modal

Empower Innovation and Creation

Fashion Design

M6 has powerful image generation and creativity capabilities, enabling the model to quickly complete the design of clothing styles and generate white background images based on user input requirements, such as "collar polo simple cardigan top."

Recommendation Reason Generation

The M6 model can capture the main features of images and convert them into textual descriptions, such as extracting product characteristics from the main product image and expanding them into a smooth and natural product recommendation reason.

Application Scenarios

E-commerce

Creation of Recommendation Copywriting

Generate Recommendations Based on Images

Supports selling point tags to enhance copy controllability

Achievable results with a small amount of samples

Image and Text Product Search

Extracts multi-modal product representation based

Supports supervised (behavioral logs) and unsupervised training modes

Integrates the latest research results in contrastive learning

Intelligent Manufacturing

Footwear, Packaging Design

Text-controlled Image Generation

Supports generating images with up to 1024*1024 resolution

Two-stage method based on VQGAN and Transformer

Car Appearance

Fusion of various car designs

Generate new styles of cars with a small amount of text or images

Multi-angle images for 3D car display

Finance

Opinion Extraction

Highly accurate financial article summarization

Accurate numerical perception