transformers

b027f8ae - Refactor GGUF loading: new converter chain & GGUFQuantizer

Refactor GGUF loading: new converter chain & GGUFQuantizer - Replace per-arch TensorProcessor classes with declarative GGUFDequantizer converters wired into the conversion-ops chain - Add GGUFQuantizer + GGUFQuantizedTensor for on-the-fly dequant via the unified loading path in core_model_loading - Simplify load_gguf_checkpoint signature; drop dummy-model state-dict pre-mapping and get_gguf_hf_weights_map - Add pytest-benchmark dep and ggml benchmark test