distillation

Modules

`modelopt.torch.puzzletron.distillation.dataset`	Real Puzzle-KD text dataset adapter for native AutoModel global KD.
`modelopt.torch.puzzletron.distillation.flash_kld`	Exact memory-bounded CE and KLD from hidden states and LM heads.
`modelopt.torch.puzzletron.distillation.global_automodel`
`modelopt.torch.puzzletron.distillation.global_kd_recipe`	Small Puzzletron extensions to NeMo AutoModel's native KD recipes.
`modelopt.torch.puzzletron.distillation.loss`
`modelopt.torch.puzzletron.distillation.synthetic_vlm`	Small deterministic image/text dataset for global-KD integration runs.
`modelopt.torch.puzzletron.distillation.tournament`	Exact-LM finalist selection and sequential native global KD.

Global distillation stage interfaces.