update app

app.py

@@ -1,82 +1,91 @@
-import gradio as gr
+import time
+from threading import Thread
 
+import gradio as gr
+import torch
+from PIL import Image
 from transformers import AutoProcessor, LlavaForConditionalGeneration
-from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer, TextIteratorStreamer
+from transformers import TextIteratorStreamer
 
-from threading import Thread
-import re
-import time 
-from PIL import Image
-import torch
 import spaces
-import requests
-
-CSS ="""
-    #component-3 {
-        height: 500px !important;
-    }"""
 
 model_id = "xtuner/llava-llama-3-8b-v1_1-transformers"
 
 processor = AutoProcessor.from_pretrained(model_id)
 
 model = LlavaForConditionalGeneration.from_pretrained(
-    model_id, 
-    torch_dtype=torch.float16, 
-    low_cpu_mem_usage=True, 
+    model_id,
+    torch_dtype=torch.float16,
+    low_cpu_mem_usage=True,
 )
 
 model.to("cuda:0")
 model.generation_config.eos_token_id = 128009
 
+
 @spaces.GPU
 def bot_streaming(message, history):
     print(message)
     if message["files"]:
-        image = message["files"][-1]["path"]
+        # 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]
+            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.")    
+            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}")
+        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": True})
-    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)
-    generated_text = ""
+
+    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.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}")
+
+    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):]
-        time.sleep(0.08)
+
+        # 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
 
 
-demo = gr.ChatInterface(fn=bot_streaming, css=CSS, fill_height=True, 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)
+demo = gr.ChatInterface(
+    fn=bot_streaming,
+    fill_height=False,
+    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
+)
 
 demo.queue(api_open=False)
-demo.launch(show_api=False, share=False)
+demo.launch(show_api=False, share=False)