ai-agents-weight-initializatio-missing-weights-in-a-pretrained-model-loaded-via-f-4594dab4

THEN When adding new layers not present in the checkpoint, either pass `_fast_init=False` to `from_pretrained` (deprecated but works) or manually re-initialize the missing weights after loading by calling the model's initialization function. Alternatively, ensure that any custom initialization logic is applied to new parameters separately after loading.

new_model = Model.from_pretrained('./checkpoint/', use_new=True, _fast_init=False)  # workaround