Refactor predict function for improved readability and update history handling

app.py

@@ -198,7 +198,12 @@ def predict(message,
         # history = state.get_prompt()[:-1]
         # logger.info(f"==== History ====\n{history}")
 
-        generation_config = dict(temperature=temperature, max_new_tokens= max_output_tokens, top_p=top_p, do_sample=do_sample, num_beams = 3, repetition_penalty=repetition_penalty)
+        generation_config = dict(temperature=temperature,
+                                 max_new_tokens=max_output_tokens, 
+                                 top_p=top_p, 
+                                 do_sample=do_sample, 
+                                 num_beams = 3, 
+                                 repetition_penalty=repetition_penalty)
 
         pixel_values = None
         if image_path is not None:
@@ -215,7 +220,12 @@ def predict(message,
 
         logger.info(f"==== Lenght Pixel values ====\n{len(pixel_values)}")
         
-        response, conv_history = model.chat(tokenizer, pixel_values, message, generation_config, history=history, return_history=True)
+        response, conv_history = model.chat(tokenizer, 
+                                            pixel_values, 
+                                            message, 
+                                            generation_config, 
+                                            history=None,
+                                            return_history=True)
         logger.info(f"==== Conv History ====\n{conv_history}")
         return response, conv_history