fix

app.py

@@ -109,6 +109,7 @@ def store_doc(doc_text: str, user_id="demo"):
     except Exception as e:
         return f"Error during storing: {e}"
 
+import traceback
 def answer(system: str, context: str, question: str, user_id="demo", history="None"):
     """UI callback: retrieve, build prompt with Qwen tags, generate answer."""
     try:
@@ -122,10 +123,13 @@ def answer(system: str, context: str, question: str, user_id="demo", history="No
         if history == "Some":
             q_vec  = embed(question).view(-1).cpu()
             store  = kb[user_id]
-            sims   = torch.matmul(store["vecs"], q_vec)          # [N]
-            if sims.shape[1] == 1:
+            vecs   = store["vecs"]
+            if vecs is None or vecs.size(0) == 0:
+                return "Knowledge base is empty or corrupted."
+            sims   = torch.matmul(vecs, q_vec)          # [N]
+            if sims.dim() > 1:
                 sims = sims.squeeze(1)
-            k      = min(4, store["vecs"].size(0))
+            k      = min(4, sims.size(0))
             idxs   = torch.topk(sims, k=k, dim=0).indices.tolist()
             context_list += [store["texts"][i] for i in idxs]
         elif history == "All":
@@ -160,7 +164,8 @@ def answer(system: str, context: str, question: str, user_id="demo", history="No
 
           
     except Exception as e:
-        return f"Error in app.py: {e}, k={k}, sims.numel()={sims.numel()}, sims.shape={sims.shape if 'q_vec' in locals() else 'N/A'}"
+        tb = traceback.format_exc()
+        return f"Error in app.py: {tb}, k={k}, sims.numel()={sims.numel()}, sims.shape={sims.shape if 'q_vec' in locals() else 'N/A'}"
     finally:
         torch.cuda.empty_cache()
 

requirements.txt

@@ -4,4 +4,5 @@ sentencepiece
 accelerate
 uvicorn
 fastapi
-pydantic
+pydantic
+traceback