Update olapp.py

olapp.py

@@ -46,14 +46,19 @@ class OlHandler(BaseHTTPRequestHandler):
             repeat_penalty=1.21
         )
 
-        answ = ""
+        #answ = ""
+
+        self.send_response(200)
+        self.send_header('Content-type','text/plain; charset=utf-8')
+        self.end_headers()
         
         for tok in generator:
             token_str = llm.detokenize([tok]).decode("utf-8", errors="ignore")
             tokens.append(tok)
             if tok == llm.token_eos():
                 break
-            answ += token_str
+            self.wfile.write(token_str.encode('utf-8'))
+            #answ += token_str
             #print(token_str, end="", flush=True)
         
         #output = llm.create_completion(
@@ -61,12 +66,9 @@ class OlHandler(BaseHTTPRequestHandler):
         #  max_tokens=32,
         #  echo=False
         #)["choices"][0]["text"]
-        
-        self.send_response(200)
-        self.send_header('Content-type','text/plain; charset=utf-8')
-        self.end_headers()
+
         #self.wfile.write(output.encode('utf-8'))
-        self.wfile.write(answ.encode('utf-8'))
+        #self.wfile.write(answ.encode('utf-8'))
         return