pub fn classify_overlay_key(key: &str) -> Option<OverlayTensor>Expand description
Classify a PEFT adapter tensor as a token-overlay tensor, or None if it is
an ordinary lora_A/lora_B weight (which super::classify_key handles).
Prefix-agnostic beyond the PEFT wrapper, matching classify_key: it accepts
both …model.embed_tokens.… and the multimodal …model.language_model. embed_tokens.… spellings. Suffix order matters — the token_adapter /
lora_embedding forms are checked before the bare .weight
(modules_to_save) form so …token_adapter.base_layer.weight is not
mis-read as a full save.