Update rag_pipeline.py

rag_pipeline.py

@@ -1,4 +1,4 @@
-# rag_pipeline.py
+# rag_pipeline.py (Debug-Version mit Indexprüfung & Logging)
 
 import os
 import pickle
@@ -12,28 +12,46 @@ from groq import Groq
 from openai import OpenAI
 import tqdm
 
-# === Load environment variables (handled by Hugging Face Secrets in deployment) ===
+print("🚀 RAG-App gestartet")
+
+# === Load environment variables (in HF Spaces über Secrets verfügbar) ===
 openai_api_key = os.getenv("OPENAI_API_KEY")
 groq_api_key = os.getenv("GROQ_API_KEY")
 
 groq_client = Groq(api_key=groq_api_key) if groq_api_key else None
 openai_client = OpenAI(api_key=openai_api_key) if openai_api_key else None
 
+# === Load SentenceTransformer model ===
+print("📦 Lade SentenceTransformer Modell...")
+model = SentenceTransformer("Sahajtomar/German-semantic")
+print("✅ Modell geladen")
 
-if not os.path.exists("faiss/faiss_index.index"):
-    raise FileNotFoundError("Fehlende FAISS-Datei: faiss_index.index")
+# === Lade FAISS-Index und Chunk-Mapping ===
+try:
+    print("📂 Lade FAISS-Index...")
+    if not os.path.exists("faiss/faiss_index.index"):
+        raise FileNotFoundError("❌ faiss_index.index fehlt!")
 
-# === Load FAISS Index and Chunks ===
-index = faiss.read_index("faiss/faiss_index.index")
-with open("faiss/chunks_mapping.pkl", "rb") as f:
-    token_split_texts = pickle.load(f)
+    if not os.path.exists("faiss/chunks_mapping.pkl"):
+        raise FileNotFoundError("❌ chunks_mapping.pkl fehlt!")
 
-# === Load SentenceTransformer model ===
-model = SentenceTransformer("Sahajtomar/German-semantic")
-chunk_embeddings = model.encode(token_split_texts, convert_to_numpy=True)
+    index = faiss.read_index("faiss/faiss_index.index")
+    with open("faiss/chunks_mapping.pkl", "rb") as f:
+        token_split_texts = pickle.load(f)
+
+    chunk_embeddings = model.encode(token_split_texts, convert_to_numpy=True)
+    print("✅ FAISS & Embeddings geladen")
 
-# === Fit UMAP ===
-umap_transform = umap.UMAP(random_state=0, transform_seed=0).fit(chunk_embeddings)
+    # UMAP initialisieren
+    umap_transform = umap.UMAP(random_state=0, transform_seed=0).fit(chunk_embeddings)
+    print("✅ UMAP fit abgeschlossen")
+
+except Exception as e:
+    print(f"❌ Fehler beim Laden von FAISS oder Chunks: {e}")
+    index = None
+    token_split_texts = []
+    chunk_embeddings = None
+    umap_transform = None
 
 def project_embeddings(embeddings, umap_transform):
     umap_embeddings = np.empty((len(embeddings), 2))
@@ -42,6 +60,8 @@ def project_embeddings(embeddings, umap_transform):
     return umap_embeddings
 
 def retrieve(query, k=5):
+    if index is None or chunk_embeddings is None:
+        return ["Kein Index verfügbar."], [], []
     query_embedding = model.encode([query], convert_to_numpy=True)
     distances, indices = index.search(query_embedding, k)
     retrieved_texts = [token_split_texts[i] for i in indices[0]]
@@ -85,11 +105,16 @@ def ask_openai_llm(prompt):
     return response.choices[0].message.content.strip()
 
 def run_qa_pipeline(user_query, k=5):
+    print(f"🔎 Frage erhalten: {user_query}")
     retrieved_texts, _, _ = retrieve(user_query, k=k)
+    if not retrieved_texts or retrieved_texts[0] == "Kein Index verfügbar.":
+        return "❌ FAISS-Index nicht verfügbar. Bitte lade den Index hoch oder führe die Preprocessing-Pipeline aus."
+
     prompt = build_prompt(user_query, retrieved_texts)
+    print("✉️ Prompt gebaut, sende an LLMs...")
     answer_groq = ask_groq_llm(prompt)
     answer_openai = ask_openai_llm(prompt)
 
     return f"\U0001f999 Groq LLaMA 3 Antwort:\n{answer_groq}\n\n\U0001f52e OpenAI GPT-4 Antwort:\n{answer_openai}"
 
-# Optional: Für Visualisierung könnte ein UMAP-Plot integriert werden, derzeit ausgelassen für Spaces
+# Ende