Recently, LinSeer, a private domain large-scale modeling platform independently developed by H3C under the guidance of Unisoc Group, received a 4+ rating in the large-scale pre-training model compliance verification of the China Institute of Information Industry, reaching the domestic advanced level. China. This comprehensive, multi-dimensional evaluation focuses on LinSeer’s five functional modules: data management, model training, model management, model deployment, and integrated development process. It demonstrates the leading strength of H3C in the field of large-scale modeling in the private sector and will provide strong support for various industries to enter the AIGC era.
As the popularity of AIGC continues to rise, the development process of large-scale AI models is accelerating, thus creating a need for standards. In this regard, the China Academy of Information Industry, in conjunction with academia, scientific research institutes, and industry, released the Trusted Artificial Intelligence Large-Scale Model Standard System 2.0. This standard system provides a comprehensive reference for scientific evaluation of the technical capabilities and application efficiency of large-scale models. H3C participated in this evaluation and comprehensively evaluated LinSeer’s development capabilities from five evaluation indicators, demonstrating its excellent technical strength.
Data management: The evaluation focuses on the data processing and version management capabilities of large-scale models, including data cleaning, annotation, quality inspection, etc. LinSeer has shown excellent performance in data cleaning completeness and functional support. Through efficient data set management and data processing, combined with the data quality detection of the Oasis platform, it can fully support the annotation of text, image, audio, and video data.
Model training: The evaluation focuses on the ability of large-scale models to support multiple training methods, visualization, and resource optimization scheduling. Based on the Model as a Service (MaaS) architecture, H3C provides comprehensive large-scale model training and fine-tuning services to generate customized and exclusive models for customers. The results show that LinSeer fully supports multi-modal training, pre-training tasks, natural language, and programming languages, with an average incremental accuracy of 91.9% and a resource utilization rate of 90%.
Model management: The evaluation focuses on the ability of large-scale models to support model storage, version management, and log management. LinSeer’s vector storage and retrieval enables models to remember and support precise answer scenarios. The results show that LinSeer can fully support model storage capabilities such as file system management and image management, as well as version management capabilities such as metadata management, relationship maintenance, and structure management.
Model deployment: Evaluate the ability of large-scale models to support model fine-tuning, transformation, pruning and quantification. LinSeer supports various fine-tuning algorithms to flexibly meet the different data and model needs of industry customers. It also provides extensive model conversion capabilities of many types. LinSeer supports model pruning and quantization, reaching advanced levels in terms of inference latency acceleration and memory usage.
Integrated development process: The assessment focuses on independent development capabilities for large models. LinSeer is integrated with H3C’s full-stack ICT infrastructure monitoring tool to organically integrate all stages of AI large-scale model development and provide a unified development platform and tools. Help industry customers effectively activate large-scale models in the private domain, quickly build intelligent applications, and achieve “freedom of model use.”
H3C implements the AI in ALL strategy and integrates artificial intelligence into a full range of software and hardware products to achieve full-stack and full-scenario technology coverage. In addition, H3C proposed the AI for ALL industry empowerment strategy, which aims to deeply understand industry needs, integrate AI capabilities into end-to-end solutions, and provide services for partners to help intelligent upgrades in various industries.
In order to further promote artificial intelligence application innovation and industrial implementation, H3C launched the AIGC overall solution, focusing on the enabling platform, data platform, and computing power platform. This comprehensive solution fully meets the needs of users’ business scenarios and helps customers quickly build large-scale private domain models with industry focus, regional focus, data exclusiveness, and value orientation.
Post time: Sep-22-2023
