roboflow/maestro
Maestro
Features
- Fine-Tuning Frameworks - A framework for fine-tuning multimodal vision-language models like Florence-2 and PaliGemma 2 on custom datasets with a streamlined Python API and CLI.
- Streamlined Fine-Tuning Pipelines - Fine-tune multimodal models on custom datasets using a streamlined configuration and training pipeline.
- Static Dictionary Definitions - Defines all training parameters (dataset, epochs, batch size, optimizer) as a static dictionary rather than imperative code.
- Vision-Language Fine-Tunings - Fine-tuning multimodal vision-language models on custom datasets for specialized computer vision tasks using a streamlined pipeline.
- Python API - Configure fine-tuning programmatically by importing a training function and passing a configuration dictionary.
- Training - Configure fine-tuning jobs from the command line by specifying dataset, epochs, batch size, optimization strategy, and metrics.
- Training Execution CLI Commands - Start a fine-tuning job from the command line by specifying dataset, epochs, batch size, optimization strategy, and metrics.
- Pipeline Configurations - Passes a single Python dictionary through the entire training lifecycle, from setup to execution.
- Training Execution APIs - Run fine-tuning programmatically by passing a configuration dictionary to a model-specific training function.
- Unified Interfaces - Wraps distinct vision-language model architectures behind a unified fine-tuning interface.
- Dispatch Mechanisms - Routes fine-tuning jobs to model-specific training functions based on a configuration dictionary key.
