Update app.py

app.py

@@ -18,22 +18,84 @@ PLACEHOLDER = """
 </div>
 """
 
-model_id = "xtuner/llava-llama-3-8b-v1_1-transformers"
+model_id_llama3 = "xtuner/llava-llama-3-8b-v1_1-transformers"
+model_id_phi3 = "xtuner/llava-llama-3-8b-v1_1-transformers"
 
-processor = AutoProcessor.from_pretrained(model_id)
+processor = AutoProcessor.from_pretrained(model_id_llama3)
+processor = AutoProcessor.from_pretrained(model_id_phi3)
 
-model = LlavaForConditionalGeneration.from_pretrained(
-    model_id,
+model_llama3 = LlavaForConditionalGeneration.from_pretrained(
+    model_id_llama3,
     torch_dtype=torch.float16,
     low_cpu_mem_usage=True,
 )
+model_llama3.to("cuda:0")
+model_llama3.generation_config.eos_token_id = 128009
 
-model.to("cuda:0")
-model.generation_config.eos_token_id = 128009
+model_phi3 = LlavaForConditionalGeneration.from_pretrained(
+    model_id_phi3,
+    torch_dtype=torch.float16,
+    low_cpu_mem_usage=True,
+)
+model_phi3.to("cuda:0")
+model_phi3.generation_config.eos_token_id = 128009
+
+
+@spaces.GPU
+def bot_streaming_llama3(message, history):
+    print(message)
+    if message["files"]:
+        # message["files"][-1] is a Dict or just a string
+        if type(message["files"][-1]) == dict:
+            image = message["files"][-1]["path"]
+        else:
+            image = message["files"][-1]
+    else:
+        # if there's no image uploaded for this turn, look for images in the past turns
+        # kept inside tuples, take the last one
+        for hist in history:
+            if type(hist[0]) == tuple:
+                image = hist[0][0]
+    try:
+        if image is None:
+            # Handle the case where image is None
+            gr.Error("You need to upload an image for LLaVA to work.")
+    except NameError:
+        # Handle the case where 'image' is not defined at all
+        gr.Error("You need to upload an image for LLaVA to work.")
+
+    prompt = f"<|start_header_id|>user<|end_header_id|>\n\n<image>\n{message['text']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
+    # print(f"prompt: {prompt}")
+    image = Image.open(image)
+    inputs = processor(prompt, image, return_tensors='pt').to(0, torch.float16)
+
+    streamer = TextIteratorStreamer(processor, **{"skip_special_tokens": False, "skip_prompt": True})
+    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024, do_sample=False)
+
+    thread = Thread(target=model_llama3.generate, kwargs=generation_kwargs)
+    thread.start()
+
+    text_prompt = f"<|start_header_id|>user<|end_header_id|>\n\n{message['text']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
+    # print(f"text_prompt: {text_prompt}")
+
+    buffer = ""
+    time.sleep(0.5)
+    for new_text in streamer:
+        # find <|eot_id|> and remove it from the new_text
+        if "<|eot_id|>" in new_text:
+            new_text = new_text.split("<|eot_id|>")[0]
+        buffer += new_text
+
+        # generated_text_without_prompt = buffer[len(text_prompt):]
+        generated_text_without_prompt = buffer
+        # print(generated_text_without_prompt)
+        time.sleep(0.06)
+        # print(f"new_text: {generated_text_without_prompt}")
+        yield generated_text_without_prompt
 
 
 @spaces.GPU
-def bot_streaming(message, history):
+def bot_streaming_phi3(message, history):
     print(message)
     if message["files"]:
         # message["files"][-1] is a Dict or just a string
@@ -63,7 +125,7 @@ def bot_streaming(message, history):
     streamer = TextIteratorStreamer(processor, **{"skip_special_tokens": False, "skip_prompt": True})
     generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024, do_sample=False)
 
-    thread = Thread(target=model.generate, kwargs=generation_kwargs)
+    thread = Thread(target=model_phi3.generate, kwargs=generation_kwargs)
     thread.start()
 
     text_prompt = f"<|start_header_id|>user<|end_header_id|>\n\n{message['text']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
@@ -85,20 +147,55 @@ def bot_streaming(message, history):
         yield generated_text_without_prompt
 
 
-chatbot=gr.Chatbot(placeholder=PLACEHOLDER,scale=1)
-chat_input = gr.MultimodalTextbox(interactive=True, file_types=["image"], placeholder="Enter message or upload file...", show_label=False)
+#chatbot=gr.Chatbot(placeholder=PLACEHOLDER,scale=1)
+#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:
-    gr.ChatInterface(
-    fn=bot_streaming,
-    title="LLaVA Llama-3-8B",
-    examples=[{"text": "What is on the flower?", "files": ["./bee.jpg"]},
-              {"text": "How to make this pastry?", "files": ["./baklava.png"]}],
-    description="Try [LLaVA Llama-3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers). Upload an image and start chatting about it, or simply try one of the examples below. If you don't upload an image, you will receive an error.",
-    stop_btn="Stop Generation",
-    multimodal=True,
-    textbox=chat_input,
-    chatbot=chatbot,
+  with gr.Row():
+    chatbot1 = gr.Chatbot(
+    [],
+    elem_id="llama3",
+    bubble_full_width=False,
+    label='LLaVa-Llama3'
+    )
+    chatbot2 = gr.Chatbot(
+    [],
+    elem_id="phi3",
+    bubble_full_width=False,
+    label='LLaVa-Phi3'
     )
+      
+  chat_input = gr.MultimodalTextbox(interactive=True, file_types=["image"], placeholder="Enter message or upload file...", show_label=False)
+
+  gr.Examples(examples=[[{"text": "What is on the flower?", "files": ["./bee.png"]}],],
+        {"text": "How to make this pastry?", "files": ["./baklava.png"]},],
+              inputs=chat_input)
+
+  #chat_input.submit(lambda: gr.MultimodalTextbox(interactive=False), None, [chat_input]).then(bot_streaming_llama3, [chat_input, chatbot1,], [chatbot1,])
+  
+  chat_msg1 = chat_input.submit(bot_streaming_llama3, [chat_input, chatbot1,], [chatbot1,])
+  chat_msg2 = chat_input.submit(bot_streaming_phi3, [chat_input, chatbot2,], [chatbot2,])
+
+  #bot_msg1 = chat_msg1.then(bot, chatbot1, chatbot1, api_name="bot_response1")
+  #chat_msg1.then(lambda: gr.MultimodalTextbox(interactive=True), None, [chat_input])
+  #bot_msg2 = chat_msg2.then(bot, chatbot2, chatbot2, api_name="bot_response2")
+  #bot_msg2.then(lambda: gr.MultimodalTextbox(interactive=True), None, [chat_input])
+
+  chatbot1.like(print_like_dislike, None, None)
+  chatbot2.like(print_like_dislike, None, None)
+
+ 
+    #gr.ChatInterface(
+    #fn=bot_streaming_llama3,
+    #title="LLaVA Llama-3-8B",
+    #examples=[{"text": "What is on the flower?", "files": ["./bee.jpg"]},
+    #          {"text": "How to make this pastry?", "files": ["./baklava.png"]}],
+    #description="Try [LLaVA Llama-3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers). Upload an image and start chatting about it, or simply try one of the examples below. If you don't upload an image, you will receive an error.",
+    #stop_btn="Stop Generation",
+    #multimodal=True,
+    #textbox=chat_input,
+    #chatbot=chatbot,
+    #)
 
 demo.queue(api_open=False)
 demo.launch(show_api=False, share=False)