Hacktoberfest 2026: los issues que los mantenedores marcaron para octubre, abiertos y aptos para principiantes. Explorar issues de Hacktoberfest

[feat] Staged training

Abierto
#151 0 comentarios 1 reacción 0 asignados Ver en GitHub

Nadie ha tomado este issue todavía.

Evaluación

Dificultad
5/5
Tiempo estimado
Más de una semana
Aptitud para principiantes
12/100
Tipo de issue
Nueva funcionalidad
Claridad
Bastante claro
Estado de actividad
Estancado
Stack tecnológico
python

Línea de trabajo

The issue proposes a staged training feature built around Trainer and a stages:list[TrainerConfig] config argument, and it has no files or tests named yet. Start by reading the existing Trainer and TrainerConfig code and related issue #126 to choose between a new StagedTrainer and extending Trainer. Done would need a design agreed with maintainers first, since the approach is still open.

Escrito por el modelo de indexación a partir del texto del issue.

Descripción

enhancement

🧐 Problem Description

We often want to have staged training where we modify the training scheme mid-training. For example we may want to:

  • Switch the dataset
  • Change the optimization scheme (lr schedule, freeze/unfreeze some weights, etc.)
  • Change the batch configuration (batch size, sequence length, etc.)
  • Tune up performance parameters according to the new configuration (including distributed config).
  • Change the cluster size

💡 Proposed Solution

Implement a "StagedTrainer"or adapt Trainer to swap the training scheme between stages.

On the config side, add a stages:list[TrainerConfig] argument that can be used to override defaults from the parent TrainerConfig (the overriding behaviour might need some work). This can either go in TrainerConfig or in as separate StagedTrainerConfig selected dynamically (#126).

🔄 Alternatives Considered

Staged training is already possible by running multiple consecutive experiments, but it's more time-consuming and less practical.

📈 Potential Benefits

More flexibility, faster development

Lenguaje dominante
Python
Estrellas
337
Forks
52
Métricas de merge de PR
Sin PR fusionados en 30 d

Preparar el entorno

Primeros pasos

  1. Lee el issue completo y luego la guía de contribución del proyecto.
  2. Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
  3. Haz un fork del repositorio y trabaja en una rama.
  4. Abre un pull request que haga referencia al número del issue.

Más de ServiceNow/Fast-LLM

Todos los issues de ServiceNow/Fast-LLM

Issues similares

Más issues de Python

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.