Jove
Visualize
Contáctanos
JoVE
x logofacebook logolinkedin logoyoutube logo
ACERCA DE JoVE
Visión GeneralLiderazgoBlogCentro de Ayuda JoVE
AUTORES
Proceso de PublicaciónConsejo EditorialAlcance y PolíticasRevisión por ParesPreguntas FrecuentesEnviar
BIBLIOTECARIOS
TestimoniosSuscripcionesAccesoRecursosConsejo Asesor de BibliotecasPreguntas Frecuentes
INVESTIGACIÓN
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchivo
EDUCACIÓN
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualCentro de Recursos para ProfesoresSitio de Profesores
Términos y Condiciones de Uso
Política de Privacidad
Políticas

Videos de Conceptos Relacionados

Downsampling01:20

Downsampling

252
When considering a sampled sequence with zero values between sampling instants, one can replace it by taking every N-th value of the sequence. At these integer multiples of N, the original and sampled sequences coincide. This process, known as decimation, involves extracting every N-th sample from a sequence, thereby creating a more efficient sequence.
The Fourier transform of the decimated sequence reveals a combination of scaled and shifted versions of the original spectrum. This...
252
Linear Approximation in Frequency Domain01:26

Linear Approximation in Frequency Domain

131
Linear systems are characterized by two main properties: superposition and homogeneity. Superposition allows the response to multiple inputs to be the sum of the responses to each individual input. Homogeneity ensures that scaling an input by a scalar results in the response being scaled by the same scalar.
In contrast, nonlinear systems do not inherently possess these properties. However, for small deviations around an operating point, a nonlinear system can often be approximated as linear....
131
Upsampling01:22

Upsampling

309
Managing signal sampling rates is essential in digital signal processing to maintain signal integrity. A decimated signal, characterized by a reduced frequency range due to its lower sampling rate, can be upsampled by inserting zeros between each sample. This upsampling process expands the original spectrum and introduces repeated spectral replicas at intervals dictated by the new Nyquist frequency. To refine this zero-inserted sequence, it is passed through a lowpass filter with a cutoff...
309
Fineness Modulus01:19

Fineness Modulus

784
The fineness modulus (FM) of aggregate is a numerical index that measures the coarseness or fineness of the particles. It is calculated by adding the cumulative percentages of aggregate retained on each of a specified series of sieves and dividing the sum by 100.
Consider performing sieve analysis on sand through a set of ASTM sieves. The weight of aggregate retained in each sieve and pan placed at the bottom is recorded, as given in Column B of Table 1.
To determine the fineness modulus of...
784
Per-Unit Sequence Models01:26

Per-Unit Sequence Models

116
An ideal Y-Y transformer, grounded through neutral impedances, displays per-unit sequence networks akin to those of a single-phase ideal transformer when subjected to balanced positive- or negative-sequence currents. These currents do not produce neutral currents, and their associated voltage drops.
Zero-sequence currents, which are identical in magnitude and phase, generate a neutral current, resulting in voltage drops across the neutral impedance and the low-voltage winding. If the...
116
Improving Translational Accuracy02:07

Improving Translational Accuracy

2.7K
2.7K

También podría leer

Artículos Relacionados

Artículos vinculados a este trabajo por autores compartidos, revista y gráfico de citas.

Ordenar por
Same author

Closed-Loop digital therapeutics empowered by deep reinforcement learning and wearable sensing for precision orthopedic rehabilitation: a simulation-based proof-of-concept study.

Frontiers in rehabilitation sciences·2026
Same author

Iron-based anodes facilitate concurrent mercury removal and bioenergy generation in constructed wetland-microbial fuel cells.

Bioresource technology·2026
Same author

Dynamic energy-aware fixed-point linear mapping multiplier for internet of things edge devices.

Scientific reports·2026
Same author

'Fuhui 631', a novel hybrid rice restorer line with multiple resistance genes, high combining ability, and superior grain quality developed with marker‑assisted breeding.

Molecular breeding : new strategies in plant improvement·2026
Same author

A pathogen lncRNA secreted into rice sequesters a host miRNA for virulence.

Nature·2026
Same author

Rapid and Specific Dual-Mode Recognition of Cathinone Derivatives Enabled by π-Conjugation-Tuned Multinoncovalent Interactions.

Analytical chemistry·2026

Video Experimental Relacionado

Updated: Sep 10, 2025

Generation and Coherent Control of Pulsed Quantum Frequency Combs
06:42

Generation and Coherent Control of Pulsed Quantum Frequency Combs

Published on: June 8, 2018

9.1K

ModuLoRA: ajuste fino de los LLM de 2 bits en las GPU de consumo mediante la integración con los cuantizadores

Junjie Yin1, Jiahao Dong2, Yingheng Wang3

  • 1Department of Computer Science, Johns Hopkins University.

Transactions on machine learning research
|August 21, 2025
PubMed
Resumen

Introducimos ModuLoRA, un algoritmo de memoria eficiente para el ajuste fino de grandes modelos de lenguaje (LLM) utilizando precisión de 2-4 bits en una sola GPU. Este método permite un ajuste fino avanzado de baja precisión, logrando un rendimiento competitivo con un uso de memoria reducido.

Más Videos Relacionados

Characterization of Anisotropic Leaky Mode Modulators for Holovideo
09:36

Characterization of Anisotropic Leaky Mode Modulators for Holovideo

Published on: March 19, 2016

8.0K
Shaping the Amplitude and Phase of Laser Beams by Using a Phase-only Spatial Light Modulator
08:39

Shaping the Amplitude and Phase of Laser Beams by Using a Phase-only Spatial Light Modulator

Published on: January 28, 2019

9.9K

Videos de Experimentos Relacionados

Last Updated: Sep 10, 2025

Generation and Coherent Control of Pulsed Quantum Frequency Combs
06:42

Generation and Coherent Control of Pulsed Quantum Frequency Combs

Published on: June 8, 2018

9.1K
Characterization of Anisotropic Leaky Mode Modulators for Holovideo
09:36

Characterization of Anisotropic Leaky Mode Modulators for Holovideo

Published on: March 19, 2016

8.0K
Shaping the Amplitude and Phase of Laser Beams by Using a Phase-only Spatial Light Modulator
08:39

Shaping the Amplitude and Phase of Laser Beams by Using a Phase-only Spatial Light Modulator

Published on: January 28, 2019

9.9K

Área de la Ciencia:

  • Inteligencia artificial
  • Aprendizaje automático
  • Procesamiento del lenguaje natural

Sus antecedentes:

  • Los grandes modelos de lenguaje (LLM) requieren recursos computacionales sustanciales para el ajuste fino.
  • Los métodos de ajuste fino existentes a menudo requieren hardware de gama alta, lo que limita la accesibilidad.
  • La cuantización de baja precisión ofrece un camino para reducir las huellas de memoria, pero presenta desafíos de ajuste fino.

Objetivo del estudio:

  • Desarrollar un algoritmo de ajuste fino eficiente en la memoria para los LLM.
  • Para permitir el ajuste fino de los LLM con parámetros 65B en las GPU de calidad de consumo.
  • Integrar cuantificadores de peso arbitrarios con adaptación de rango bajo para un ajuste fino flexible.

Principales métodos:

  • Propuso ModuLoRA (adaptación modular de bajo rango), un nuevo enfoque de ajuste fino.
  • Implementado un paso hacia atrás agnóstico de la cuantización para la materialización de peso de baja precisión adaptativa.
  • Métodos de cuantización integrados de 2 bits QuIP# y 3 bits OPTQ.

Principales resultados:

  • LLM ajustados con éxito con parámetros de 65B utilizando precisión de 2/3/4 bits en una sola GPU de 24 GB.
  • Logró un rendimiento competitivo en la clasificación de textos, la inferencia del lenguaje natural y las tareas de seguimiento de instrucciones.
  • Superó las puntuaciones ROUGE de última generación en las tareas de resumen, superando los métodos existentes de 4 y 8 bits.

Conclusiones:

  • ModuLoRA reduce significativamente los requisitos de memoria para el ajuste fino de LLM.
  • El método permite por primera vez el ajuste fino de LLM de alta precisión (2 bits, 3 bits).
  • Lanzado ModuLoRA y modelos de baja precisión a través de la biblioteca LLMTools para una mayor accesibilidad.