faz7_rag.py: --int8 dynamic quant (~2x kucuk bellek 708->325MB, ~1.2x hiz, ciktilar birebir)

faz7_rag.py: decode-cache (step/KV-cache) — CPU uretim O(L), ~6x hizli, full-recompute ile birebir

faz7_rag: --device cpu (LambaCPU saf-PyTorch yolu; fork import soft, CPU RAG yerelde calisiyor)

LAMBA release stage: faz7_rag.py

RAG-SFT: faz7_rag.py (extractive QA converter + prompt hizalama)

faz7_rag: kisa/no-think talimati + <think> strip + --max_new 96

faz7_rag: RAG prompt SFT formatina hizalandi + dil-duyarli yonerge

faz7_rag: --ab modu (RAG vs ham, tek kosu) + DEMO_QUERIES

faz7_rag: RAG pipeline (retrieve+compress+inject, demo corpus)