add timeout 3 min

app.py

@@ -6,6 +6,7 @@ from langchain_community.document_loaders import PyPDFLoader
 from langchain_community.vectorstores import FAISS
 from langchain_huggingface.embeddings import HuggingFaceEmbeddings
 from langchain.text_splitter import RecursiveCharacterTextSplitter
+import concurrent.futures
 
 # ── KONFIG ───────────────────────────────────────────────────────────
 DOCS_DIR       = "document"
@@ -31,7 +32,8 @@ log = logging.getLogger(__name__)
 
 # ── 1) Index (bygg eller ladda) ─────────────────────────────────────
 emb = HuggingFaceEmbeddings(model_name=EMB_MODEL)
-if os.path.isdir(INDEX_DIR):
+INDEX_PATH = os.path.join(INDEX_DIR, "index.faiss")
+if os.path.isfile(INDEX_PATH):
     log.info(f"🔄 Laddar index från {INDEX_DIR}")
     vs = FAISS.load_local(INDEX_DIR, emb)
 else:
@@ -45,7 +47,6 @@ else:
             docs.extend(chunks); pdfs.append(fn)
     vs = FAISS.from_documents(docs, emb); vs.save_local(INDEX_DIR)
     log.info(f"✅ Byggt index – {len(pdfs)} PDF / {len(docs)} chunkar")
-
 retriever = vs.as_retriever(search_kwargs={"k": K})
 
 # ── 2) LLM‑pipeline & tokenizer ─────────────────────────────────────
@@ -86,18 +87,15 @@ def chat_fn(q, temp, max_new_tokens, k, ctx_tok_max, history):
     history = history or []
     history.append({"role": "user", "content": q})
 
-    # Skapa en ny retriever med valt k
     retriever = vs.as_retriever(search_kwargs={"k": int(k)})
     docs = retriever.invoke(q)
     if not docs:
         history.append({"role": "assistant", "content": "🚫 Hittade inget relevant."})
         return history, history
 
-    # Visa chunkar som används
     chunk_info = "\n\n".join([f"{i+1}. ({d.metadata['source']}) {d.page_content[:160]}…" for i, d in enumerate(docs)])
     history.append({"role": "system", "content": f"🔎 Chunkar som används:\n{chunk_info}"})
 
-    # Bygg prompt med valt ctx_tok_max
     def build_prompt_dynamic(query, docs, ctx_tok_max):
         context_parts = []
         total_ctx_tok = 0
@@ -120,8 +118,8 @@ def chat_fn(q, temp, max_new_tokens, k, ctx_tok_max, history):
     prompt = build_prompt_dynamic(q, docs, ctx_tok_max)
     log.info(f"Prompt tokens={len(tokenizer.encode(prompt))}  temp={temp}  max_new_tokens={max_new_tokens} k={k} ctx_tok_max={ctx_tok_max}")
 
-    try:
-        ans = gen_pipe(
+    def generate():
+        return gen_pipe(
             prompt,
             temperature=float(temp),
             max_new_tokens=int(max_new_tokens),
@@ -130,6 +128,13 @@ def chat_fn(q, temp, max_new_tokens, k, ctx_tok_max, history):
             do_sample=True,
             return_full_text=False
         )[0]["generated_text"]
+
+    try:
+        with concurrent.futures.ThreadPoolExecutor() as executor:
+            future = executor.submit(generate)
+            ans = future.result(timeout=180)  # Timeout in seconds
+    except concurrent.futures.TimeoutError:
+        ans = "⏰ Ingen respons från modellen inom 3 minuter."
     except Exception as e:
         log.exception("Genererings‑fel")
         ans = f"❌ Fel vid generering: {type(e).__name__}: {e}\n\nPrompt:\n{prompt}"