Update app.py

app.py

@@ -125,34 +125,34 @@ def predict_image(image_url, text, file_pref):
                             ]
                         }
                     ],
-                    # temperature=0.7,
-                    # max_tokens=4096,
-                    # top_p=1,
-                    # stream=True
-
-                    temperature=1,
-                    max_completion_tokens=4096,
+                    temperature=0.7,
+                    max_tokens=4096,
                     top_p=1,
-                    stream=False,
-                    stop=None,
+                    stream=True
+
+                    # temperature=1,
+                    # max_completion_tokens=4096,
+                    # top_p=1,
+                    # stream=False,
+                    # stop=None,
                 )
                 print(f"Completions: {completion}")
-                # for chunk in completion:
-                #     if chunk.choices[0].delta.content:
-                #         streamer.put(chunk.choices[0].delta.content)
-                # streamer.put(None)  # Signal the end
+                for chunk in completion:
+                    if chunk.choices[0].delta.content:
+                        streamer.put(chunk.choices[0].delta.content)
+                streamer.put(None)  # Signal the end
             except Exception as e:
                 print(f"Error in generate_response: {e}")
                 streamer.put(None)
 
-        # thread = Thread(target=generate_response)
-        # thread.start()
+        thread = Thread(target=generate_response)
+        thread.start()
         
-        # buffer = ""
-        # for new_text in streamer:
-        #     buffer += new_text
-            buffer = completion.choices[0].message
-            print(buffer)
+        buffer = ""
+        for new_text in streamer:
+            buffer += new_text
+            # buffer = completion.choices[0].message
+            # print(buffer)
         return buffer
         
     except Exception as e: