Update app.py

app.py

@@ -36,6 +36,8 @@ PLACEHOLDER = """
 def bot_streaming(message, history):
     print(f'message is - {message}')
     print(f'history is - {history}')
+    
+    image = None
     if message["files"]:
         if type(message["files"][-1]) == dict:
             image = message["files"][-1]["path"]
@@ -45,10 +47,8 @@ def bot_streaming(message, history):
         for hist in history:
             if type(hist[0]) == tuple:
                 image = hist[0][0]
-    try:
-        if image is None:
-            raise gr.Error("You need to upload an image for Phi3-Vision to work. Close the error and try again with an Image.")
-    except NameError:
+    
+    if image is None:
         raise gr.Error("You need to upload an image for Phi3-Vision to work. Close the error and try again with an Image.")
 
     conversation = []
@@ -58,21 +58,22 @@ def bot_streaming(message, history):
             flag = True
             conversation.extend([{"role": "user", "content": ""}])
             continue
-        if flag == True:
-            conversation[0]['content'] = f"\n{user}"
+        if flag:
+            conversation[0]['content'] = f"<|image|>\n{user}"
             conversation.extend([{"role": "assistant", "content": assistant}])
             flag = False
             continue
         conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
 
     if len(history) == 0:
-        conversation.append({"role": "user", "content": f"\n{message['text']}"})
+        conversation.append({"role": "user", "content": f"<|image|>\n{message['text']}"})
     else:
         conversation.append({"role": "user", "content": message['text']})
+    
     print(f"prompt is -\n{conversation}")
     prompt = processor.tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True)
     image = Image.open(image)
-    inputs = processor(prompt, image, return_tensors="pt").to("cuda:0")
+    inputs = processor(prompt, images=image, return_tensors="pt").to("cuda:0")
 
     # Custom prompt to ensure responses are in Arnold's style
     system_prompt = (
@@ -82,10 +83,10 @@ def bot_streaming(message, history):
         "'What are you doing? This is no time for games! Upload a real exercise picture and let's pump it up!'"
     )
 
-    streamer = TextIteratorStreamer(processor, **{"skip_special_tokens": True, "skip_prompt": True, 'clean_up_tokenization_spaces': False,})
-    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=280, do_sample=False, temperature=0.0, eos_token_id=processor.tokenizer.eos_token_id,)
+    streamer = TextIteratorStreamer(processor, skip_special_tokens=True, skip_prompt=True, clean_up_tokenization_spaces=False)
+    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=280, do_sample=False, temperature=0.0, eos_token_id=processor.tokenizer.eos_token_id)
 
-    thread = Thread(target=model.generate, kwargs=generation_kwargs, args=(system_prompt,))
+    thread = Thread(target=model.generate, kwargs=generation_kwargs)
     thread.start()
 
     buffer = ""
@@ -94,6 +95,7 @@ def bot_streaming(message, history):
         yield buffer
 
 
+
 chatbot = gr.Chatbot(scale=1, placeholder=PLACEHOLDER)
 chat_input = gr.MultimodalTextbox(interactive=True, file_types=["image"], placeholder="Enter message or upload file...", show_label=False)
 with gr.Blocks(fill_height=True,) as demo: