TensorNameMapper

Abstracts the mapping from logical tensor roles to GGUF tensor name strings.

Different model families (LLaMA, Gemma, Mistral, Qwen) share the same transformer architecture but use different naming conventions for their weight tensors. This interface provides a single point of indirection so that a weight loader can work with any naming scheme.

Implement this interface for each model family and pass it to the weight loader to enable loading models with non-standard tensor names.

Inheritors