Fix streaming

app.py

@@ -39,19 +39,17 @@ def generate_text(prompt):
     output_ids = []
     streamer_output = ""
 
-    def token_streamer():
-        for token_id in current_model.generate(
-            **inputs,
-            max_new_tokens=256,
-            do_sample=False,
-            return_dict_in_generate=True,
-            output_scores=False
+    for token_id in current_model.generate(
+        **inputs,
+        max_new_tokens=256,
+        do_sample=False,
+        return_dict_in_generate=True,
+        output_scores=False
         ).sequences[0]:
-            output_ids.append(token_id.item())
-            yield current_tokenizer.decode(output_ids, skip_special_tokens=True)
+        
+        output_ids.append(token_id.item())
 
-    for partial_output in token_streamer():
-        yield partial_output
+        yield current_tokenizer.decode(output_ids, skip_special_tokens=True)
 
 
 # Model options