tools

Modules

`modelopt.torch.puzzletron.tools.bypassed_training`	Utilities for initializing child models from parent models via bypassed training.
`modelopt.torch.puzzletron.tools.checkpoint_utils`	Utilities for loading and initializing PyTorch model checkpoints (AnyModel / HF layouts).
`modelopt.torch.puzzletron.tools.checkpoint_utils_hf`	Utilities for loading and saving Hugging Face-format checkpoints (`AutoConfig` + optional `block_configs`).
`modelopt.torch.puzzletron.tools.common`
`modelopt.torch.puzzletron.tools.hydra_utils`	Utilities for hydra config initialization.
`modelopt.torch.puzzletron.tools.kd_model`	Knowledge distillation loss functions.
`modelopt.torch.puzzletron.tools.logger`
`modelopt.torch.puzzletron.tools.sharded_checkpoint_utils`	Provides utilities for distributed loading, saving, and manipulation of large language model checkpoints across multiple GPUs/processes.
`modelopt.torch.puzzletron.tools.validate_model`	Checkpoint validation through the native AutoModel scoring recipe.
`modelopt.torch.puzzletron.tools.validate_puzzle_with_multi_replacements`	Validates puzzle solutions by applying layer replacements and evaluating model performance.
`modelopt.torch.puzzletron.tools.validation_utils`	Stable validation-artifact writer shared by AutoModel scoring paths.

Shared tools: logging, hydra config, checkpoint utilities, and validation helpers.