TRS DT-AI Platform

The AI development platform is an important hub connecting the upstream and downstream industries, and is also an important component of the new AI infrastructure. It provides full-process services of traditional machine learning and deep learning development, deployment, application, and operation and maintenance, which can meet the needs of different types of developers such as algorithm engineers, data scientists, and industry experts, and effectively promote the empowerment of artificial intelligence in all walks of life.

TRS DT-AI platform focuses on the three core technologies of natural language understanding (NLP), knowledge graphs, and multi-modal retrieval, combined with robotic process automation (RPA), data Middle Platform and other technologies, and is oriented to media convergence, financial technology, intelligence patents, and smart technology. Application scenarios such as public security, smart government affairs, and open source intelligence analysis cover the training, fine-tuning, compression, deployment, inference, and monitoring of large + small models as well as the full life cycle process, providing full-stack AI such as text, audio and video, and multi-modality Service capabilities can quickly and efficiently develop, train and deploy machine learning models and deep learning models of any scale, opening up the last mile of enterprise-level "large + small" model applications.

Product Advantages

Complete system and complete functions

Focusing on the three core technologies of NLP, knowledge graph, and multi-modal retrieval, it integrates a variety of algorithm models that have been trained and tuned to provide users with high-quality AI service; it provides machine learning and deep learning for text, images, audio and video, etc. There are more than 300 large + small models of different algorithms. Provide full-stack AI technology capabilities, provide hundreds of calling interfaces, and provide SDK development mode.

Scene landing

The AI+ strategy fully unleashes the power of big data by integrating it with industry data such as media, finance, public opinion and security, and realizes the application model of "AI+ industry scenarios" in various industries.
　　

Can be privatized and customized

Supports privatized deployment to ensure data security for industry users; provides training platform and training interfaces, and can customize annotated data sets and develop industry models for the field.
　　

Xinchuang software and hardware support

Support Xinchuang (information technology application innovation) software and hardware products, including OS such as Kirin, Tongxin, and Deepin, domestic CPUs such as Phytium, Kunpeng, and Loongson, and GPU products such as Huawei Ascend and Hygon DCU. It supports integrating hardware of different architectures such as Intel, Nvidia,Phytium and Loongson into a unified computing service framework.

Product Functions

One-stop AI development

Build a one-stop AI capability platform around industry applications, providing full life cycle management functions for AI development, including data annotation, model design, training, optimization, evaluation, release, download and other complete processes; providing a full-process graphical operation interface, It supports visual modeling and Notebook modeling. Users can complete algorithm model training and use without programming or low-code programming, lowering the development threshold and improving development efficiency. The generated AI model can be quickly released as an AI service, effectively improving the industry application effects of related products.

Intelligent data annotation

In response to the problem of a large amount of high-quality annotated data required for AI training, the data annotation platform can provide users with intelligent annotation services for multi-modal data such as text, images, audio and video, signals, etc., including automatic Pre-annotation based on LLM., visual annotation, annotation quality assessment and other functions. Users can build their own business annotation data sets and generate exclusive industry AI models.

Large + small model optimization

Supports AutoML and provides optimization management of deep learning models and traditional machine learning, including parameter tuning settings, neural network model compression, incremental training and other functions.

Supports fast training and tuning of large + small models, supports full parameter tuning, and supports Parameter-Efficient Fine-Tuning of Lora, qLora etc.

knowledge driven

The main driving force of deep learning comes from manually annotated big data and high-performance computing power. As high-quality annotated data, knowledge graphs are also an important information source for machine learning. Apply knowledge resources such as industry knowledge graphs to deep learning, add big data-driven deep neural networks to knowledge-driven, and improve cognitive computing effects through integrated learning of multiple models. On the other hand, deep learning results are used for knowledge discovery, forming a virtuous cycle.

AI Ability

Natural language processing:

[Knowledge Processing] provides basic text mining capabilities, covering underlying capabilities of different granularities such as words, phrases, sentences, articles, etc., and discovers the value of its existence from the text.

[Knowledge Discovery] Provides text classification, clustering, topic analysis, sentiment analysis, event identification and extraction, etc. to discover hidden value from text. Various models support customization according to industry characteristics.

Language generation:

Based on large models and knowledge graphs, combined with RAG and other technologies, automatic language generation is realized, including automatic news writing, intelligent question and answer, short title generation, report generation and other language generation in various application scenarios.

OCR

Based on a variety of algorithms, it detects and recognizes text in different scenarios. It is suitable for multiple scenarios such as taking pictures, scanning documents, handwriting, and natural scenes. It also provides text recognition for different lengths, different fonts, and different languages.

Content identification:

Based on multi-modal analysis technology such as text, images, audio and video, it can automatically detect content related to pornography, advertising, terrorism and violence, and sensitive persons, helping customers reduce the risk of business violations.

Image recognition:

Based on deep learning algorithms, it accurately identifies important information contained in images and can be applied to image classification, target detection, scene recognition and other scenarios. It can also provide image recognition in professional fields by combining industry knowledge.

Image and video search:

Based on technologies such as image semantics and image feature extraction and analysis based on multi-modal LLM, deep genetic coding calculations are performed on images, and search technologies such as image-text hybrid indexing are used to combine different application businesses and industry scenarios to help customers extract information from the image gallery. Search for the picture you need.

Machine translation:

The deep neural translation model based on the encoder and decoder architecture uses transformer, multi-feature fusion and other technologies to generate smooth translations for a variety of language pairs, faithfully express the original content, and meet high-quality translation requirements.

Application Scenario

Machine Learning Platform for Postal Savings Bank of China

Shenzhen International Innovation Industry Information Service Platform Project

China Southern Power Grid AI Platform

Union Intelligent Biomedical Literature Integrated Service Platform

By building an machine learning platform, we can build an environment and capabilities to conveniently develop enterprise-level machine learning applications, and support full-process machine learning services from data to applications. The platform focuses on unstructured data assets such as documents as the main processing objects, with business scenarios as the center, providing full-process support for natural language capability combination and service implementation, integrating annotation tools and model development tools, and providing artificial intelligence such as machine learning and deep learning. Intelligent model development support and services meet the bank's needs for analysis and processing of unstructured data such as documents and texts.

This platform benchmarks against the world's first-class standards, builds an intelligent analysis engine with relevant knowledge extraction and efficient expert collaboration as its core, builds a modular and knowledge-based innovation industry application service matrix that covers the heterogeneous needs of multiple users, and provides precise policy implementation for the government's innovation industry. , efficient transformation of scientific and technological achievements and open innovation of enterprises to provide world-class innovative industry information services. TRS DT-AI Platform provides two major types of services: algorithm library management and model algorithms related to intelligent analysis engines for national innovation projects. Algorithm library management includes data annotation, algorithm editing, algorithm training, model management, etc. Model algorithms include intelligent perception algorithms, expert collaboration algorithms, knowledge graph algorithms, data processing algorithms, evaluation algorithms, etc.

TRS DT-AI Platform provides NLP training platform based on the deep learning framework, providing users with one-stop model training and research and development capabilities, solving the problems of previous AI applications such as lack of computing resources, insufficient algorithm accumulation, and scattered datasets. Specifically, it includes providing dozens of core NLP components to provide users with intelligent data analysis ability; building a semantic model customization and optimization platform for the power industry, building its own industry annotation data through intelligent annotation tools, and providing intelligent training services through NLP models Realize the development of own AI models.

TRS DT-AI Platform provides model training tools based on deep learning frameworks, providing users with one-stop model training and R&D capabilities such as model construction and evaluation, solving problems such as analytical model construction and intelligent annotation of medical scientific literatures, and assisting medical information research.