transformers

feat(quantizers): support non-weight param names in TorchAo safetensors loading

#46325

Open