Bibliographic citations
Salgado, J., (2024). Diseño de una plataforma tecnológica de alto rendimiento con procesamiento por gráfica y basada en tecnología de hiperconvergencia para la aplicación de la inteligencia generativa en el desarrollo de sistemas predictivos de información con el entrenamiento de modelos de Inteligencia Artificial (IA) [Trabajo de Suficiencia Profesional, Universidad Peruana de Ciencias Aplicadas (UPC)]. http://hdl.handle.net/10757/674703
Salgado, J., Diseño de una plataforma tecnológica de alto rendimiento con procesamiento por gráfica y basada en tecnología de hiperconvergencia para la aplicación de la inteligencia generativa en el desarrollo de sistemas predictivos de información con el entrenamiento de modelos de Inteligencia Artificial (IA) [Trabajo de Suficiencia Profesional]. PE: Universidad Peruana de Ciencias Aplicadas (UPC); 2024. http://hdl.handle.net/10757/674703
@misc{renati/419524,
title = "Diseño de una plataforma tecnológica de alto rendimiento con procesamiento por gráfica y basada en tecnología de hiperconvergencia para la aplicación de la inteligencia generativa en el desarrollo de sistemas predictivos de información con el entrenamiento de modelos de Inteligencia Artificial (IA)",
author = "Salgado Paraguay, Julio Cesar",
publisher = "Universidad Peruana de Ciencias Aplicadas (UPC)",
year = "2024"
}
With the advances currently made in the field of Deep Learning, we seek to formulate the conceptual design of a technological platform based on hyperconverged technology, which incorporates graphics processing infrastructure or GPU (Graphics Processing Unit), capable of supporting the development of RAG (Retrieval Augmented Generation) type solutions with the application of large open source language generative models, which have been previously trained with billions of data, for coding of predictive information systems (PIS) and that also has the training of proprietary data generated within the same institution, coming from different sources of information, to optimize the precision of the information, which can accompany the innovation processes and entrepreneurship in order to increase the productivity of the productive chains of the national sector, thereby generating economic benefits that have an impact on the improvement of the quality of life of Peruvians. Thus, with the development of the project, the VRAM bandwidth will increase with information transfer rates equal to 864GB/s and 2TB/s; also, the number of floating point operations per second will be enhanced with processing rates of 378 TFLOPS in TF32 precision, 756 TFLOPS in BFLOAT16 format, 756 TFLOPS in FP16 precision and 1513 TFLOPS in FP8 format; in addition, the clock speed of the computing cores will be raised to 1755MHz, taking advantage of the capabilities of parallel processing by incorporating NVIDIA NVLink technology; likewise, the number of tokens in the context window will be increased to achieve maximums of 7,993 and 9,193 tokens/s in the output inference and, finally, the number of parameters or internal variables will be increased with the possibility of processing language models 7B, 13B, 70B, 130B, among others.
This item is licensed under a Creative Commons License