Digital Transformation Strategy

Digital transformation strategy

Transformation goals

Intelligent equipment management

By introducing the Internet of Things technology, we can achieve real-time monitoring and predictive maintenance of equipment.

Green and low-carbon development

Promote the application of new energy technologies and achieve continuous optimization of carbon emissions.

Efficient data analysis

Build an integrated data analysis platform, optimize resource allocation, and improve decision-making accuracy. Collaborative work platform: Deploy modern collaboration tools to improve cross-departmental communication efficiency and promote business agility.

提升对话总结质量的Qwen1.5-MoE-A2.7B模型

为了应对对话总结中的挑战,如对话冗长、角色互动和概念转变,我们引入了Qwen1.5-MoE-A2.7B模型。这款基于MoE(Mixture-Of-Experts)的改进模型通过在不显著增加计算预算的情况下使用大量参数,显著提升了对话总结的效果。 Qwen1.5-MoE-A2.7B模型结合了QLora和FlashAttention技术,以优化训练过程。QLora通过量化低秩适配器来减少内存使用,而FlashAttention-2则提升了训练速度和效率。此外,使用噪声嵌入指令微调(NEFTune)增强了模型的鲁棒性。这些技术使Qwen1.5-MoE-A2.7B在SAMSUM和CSDS对话总结数据集上取得了领先的性能。 在对话总结领域,传统的小型语言模型如BERT和BART虽然表现良好,但大规模模型通常能带来显著的性能提升。最新的基于LLM的模型如DIALOGLM和GatorTronGPT-20B虽然在对话理解上表现出色,但Qwen1.5-MoE-A2.7B通过细粒度专家和共享专家等创新,达到了与当前最先进模型相当的效果。 通过这些先进的技术,Qwen1.5-MoE-A2.7B显著提升了对话总结的质量和效率,为公司在自然语言处理领域提供了强有力的工具。

精调多模态大语言模型用于X光图像判断

本研究开发了一种多模态深度学习模型,针对医学影像问答(QA)和癌症诊断任务。通过构建指令问答数据集,并对Qwen-VL 7B模型进行微调,模型在Rouge评分上显著优于其他多模态模型。

基于DenseNet模型进行癌症图像分类

探讨了如何利用DenseNet模型进行癌症图像分类。DenseNet是一种创新的深度学习架构,通过密集连接的网络层次结构,显著提升了模型的性能。该研究专注于应用于转移癌图像分类任务,利用PatchCamelyon (PCam) 数据集进行实验。通过对比不同模型,研究表明DenseNet模型在AUC-ROC和准确率方面均优于传统的ResNet34和VGG19模型,特别是DenseNet201 (TTA)模型在AUC-ROC评分上提升了2.37%,在准确率上提高了2.4%。这些结果表明,DenseNet在癌症图像分析中的应用表现出色,同时数据增强技术也显著提升了模型的分类能力。未来的研究将继续优化DenseNet模型,以进一步提高其在癌症图像分类任务中的效果。

Strategic Pillars

Technology empowers production

Through advanced technologies such as the Internet of Things and big data analysis, we can achieve the transformation from traditional models to intelligent production.

Ecological synergy upgrade

Promote deep synergy between upstream and downstream industrial chains to improve overall efficiency and competitiveness.

Innovation drives growth

Build an innovation institute, strengthen technology research and development and achievement transformation, and consolidate the leading position in the industry.

Talent and cultural support

Strengthen the digital capability building of employees and create an innovation-driven corporate culture atmosphere.

Project Vision

Build a domestic leading intelligent and green aluminum enterprise, realize the comprehensive transformation and upgrading from traditional manufacturing to digitalization and ecology, and provide a demonstration for achieving sustainable development and rural revitalization.