TRACE: TRansformer-based Attribution using Contrastive Embeddings in LLMs