---

name: training description: "Route MambaVision ImageNet training and fine-tuning workflows." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Training

Use this sub-skill for MambaVision ImageNet training and fine-tuning command planning.

It covers:

• choosing among the bundled YAML presets for mamba_vision_T, mamba_vision_T2, mamba_vision_S, mamba_vision_B, mamba_vision_L, and mamba_vision_L2
• adapting the published torchrun launch pattern into safe command recipes with user-supplied paths
• train.py flags for data roots, split names, model choice, input size, batch size, AMP, EMA, MESA, checkpoints, logging, and output locations
• ImageFolder and LMDB-style data layouts
• distributed vs single-GPU debugging and failure triage
• validation that runs inside the training loop

Use this sub-skill only for training and fine-tuning. Route other tasks elsewhere:

• classification API, inference, feature extraction, or standalone validation -> ../classification/SKILL.md
• OpenMMLab detection training -> ../object-detection/SKILL.md
• OpenMMLab segmentation training -> ../semantic-segmentation/SKILL.md

Start with the bundled references and helper:

• references/training-workflows.md for launch recipes and flag choices.
• references/configuration.md for YAML preset differences.
• references/data-formats.md for ImageNet/ImageFolder/LMDB layout checks.
• references/troubleshooting.md for OOM, DDP, data, checkpoint, and numerical failures.
• scripts/print_training_command.py to print a safe command template without launching training.

This skill is a router and reference hub. It does not bundle or launch the full training entry point because full ImageNet training is long-running and target-project specific.