SAFE-KV (Findings of AACL-IJCNLP 2026): causal suffix repair makes non-prefix KV-cache reuse in RAG reproduce full-prefill answers. One-file implementation, plus the code, data, and per-example outputs behind every number in the paper.
nlp reproducible-research pytorch transformer reproducibility efficient-inference rag prefill hotpotqa kv-cache large-language-models llm efficient-nlp prompt-caching llm-inference retrieval-augmented-generation 2wikimultihopqa multi-hop-qa kv-cache-reuse aacl-ijcnlp
-
Updated
Sep 24, 2026 - Python