Trainers registry

Trainer classes selected by train.type. See Architecture and extension seams.

Build any of these from a config with train = dict(type=...), using the type string in the first column. Generated from the live registry - 5 classes (5 names including aliases), grouped by role.

General

The default training loop and multi-dataset variant.

Name	Summary
DefaultTrainer	Default single-dataset supervised/SSL trainer.
MultiDatasetTrainer	Trainer that draws mixed batches from several datasets.

pimm.engines.train.Trainer	Default single-dataset supervised/SSL trainer.
pimm.engines.train.MultiDatasetTrainer	Trainer that draws mixed batches from several datasets.

Task-specific

Trainers with task-specific collation / loaders.

Name	Summary
ImageClassTrainer	Trainer for dense 2D image batches (e.g. rasterized ring images).
InsegTrainer	Trainer for instance segmentation with instance-aware collation.

pimm.engines.train.ImageClassTrainer	Trainer for dense 2D image batches (e.g. rasterized ring images).
pimm.engines.train.InsegTrainer	Trainer for instance segmentation with instance-aware collation.

Reinforcement learning

Rollout-based policy optimization (GRPO).

Name	Summary
GRPOTrainer	Reinforcement-learning trainer implementing GRPO at the trainer level.

pimm.engines.train.GRPOTrainer

Reinforcement-learning trainer implementing GRPO at the trainer level.