Spaces: lazy LLM init + filepath upload

day3/rag_system.py

@@ -1,31 +1,41 @@
-# rag_system.py
+# day3/rag_system.py
 from typing import List, Dict
+import os
 import chromadb
+
 from pdf_loader import load_pdf
 from optimal_chunker import OptimalChunker
 from embeddings import embed_texts
+
+# LLM-i sonradan yaradacağıq
 from langchain_groq import ChatGroq
 from dotenv import load_dotenv
-import os
 
 load_dotenv()
 
 
 class RAGPipeline:
-    def __init__(self, persist_dir: str = "./chroma_db", collection_name: str = "pdf_docs"):
-        # Vector DB (Chroma 1.x new API)
+    def __init__(self, persist_dir: str = "./chroma_db_space", collection_name: str = "pdf_docs"):
+        # Vector DB (Chroma 1.x)
         self.client = chromadb.PersistentClient(path=persist_dir)
         self.col = self.client.get_or_create_collection(name=collection_name)
 
         # Chunker
         self.chunker = OptimalChunker()
 
-        # LLM (Groq)
-        self.llm = ChatGroq(
-            model="llama3-8b-8192",
-            temperature=0.0,
-            api_key=os.getenv("GROQ_API_KEY"),
-        )
+        # LLM hələ YARADILMIR (lazy)
+        self.llm = None
+
+    def _ensure_llm(self):
+        """GROQ_API_KEY varsa LLM-i gec (ilk sorğuda) yarat."""
+        if self.llm is None:
+            api_key = os.getenv("GROQ_API_KEY")
+            if not api_key:
+                # LLM olmadan da app işə düşsün deyə aydın mesaj veririk
+                raise RuntimeError(
+                    "GROQ_API_KEY tapılmadı. Space Settings → Variables and secrets bölməsində əlavə edin."
+                )
+            self.llm = ChatGroq(model="llama3-8b-8192", temperature=0.0, api_key=api_key)
 
     # 1) Load  2) Chunk  3) Embed  4) Upsert to Chroma
     def index_document(self, pdf_path: str, doc_id_prefix: str = "doc") -> Dict:
@@ -35,8 +45,7 @@ class RAGPipeline:
         summary = self.chunker.fit_on_text(text)
         chunks = self.chunker.transform()
 
-        # embeddings: list[list[float]]
-        vectors = embed_texts(chunks)
+        vectors = embed_texts(chunks)  # list[list[float]]
         ids = [f"{doc_id_prefix}-{i}" for i in range(len(chunks))]
 
         self.col.add(
@@ -49,6 +58,9 @@ class RAGPipeline:
 
     # 5) Retrieve  6) Ask LLM
     def query(self, question: str, k: int = 4) -> Dict:
+        # LLM-i bu zaman yaradacağıq (secret yoxdursa burda aydın xəta görünəcək)
+        self._ensure_llm()
+
         results = self.col.query(query_texts=[question], n_results=k)
         chunks: List[str] = results["documents"][0] if results.get("documents") else []
 
@@ -66,24 +78,9 @@ Context:
 
 Answer (verbatim from Context):"""
         resp = self.llm.invoke(prompt)
-        answer = resp.content.strip()
+        answer = resp.content.strip() if hasattr(resp, "content") else str(resp)
 
-        # Fallback if the model still hedges
         if (not answer or answer.lower().startswith("i don't know")) and context.strip():
             answer = chunks[0] if chunks else "I don't know"
 
-        return {
-            "answer": answer,
-            "used_chunks": len(chunks),
-            "context_preview": context[:500],
-        }
-
-
-if __name__ == "__main__":
-    rag = RAGPipeline()
-    info = rag.index_document("sample.pdf")  # ensure day3/sample.pdf exists
-    print("Indexed:", info)
-
-    out = rag.query("What text does the PDF contain?")
-    print("Answer:", out["answer"])
-    print("Used chunks:", out["used_chunks"])
+        return {"answer": answer, "used_chunks": len(chunks), "context_preview": context[:500]}