Missing definition - checkpointing

[madragonse]

    model_config = ConfigDict(extra="forbid")
    learning_rate: PositiveFloat
    weight_decay: PositiveFloat
    scheduler: object
    gradient_accumulation_steps: PositiveInt
    n_steps: PositiveInt
    gradient_clipping: PositiveFloat
    evaluation: dict
    # TODO missing checkpoint

# TODO missing checkpoint class
# EXAMPLE: 
# save:
#     interval: 1000000
#     model_checkpoint_filename: __model_optim_scheduler__.pt
#     training_state_filename: __training_state__.pt
#     path: checkpoints # Change me!

#    load:
#     type: nano
#     path: "/home/mstefaniak/llmrandom_cemetery/run_exp_2025-07-28_11-00-53/checkpoints/0/step_-1"
#     model_checkpoint_filename: "__training_state__.pt"```

[madragonse]

Do uses default values - shouldn't.

[crewtool]

@madragonse could you elaborate on that?

[madragonse]

general lack of definitions i know now, it referes to all configs to be evaled according to to-be-done in the future class definitions, prefereable with pydantic eval