Update app.py

app.py

@@ -1,54 +1,73 @@
+import os
 import torch
 import gradio as gr
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-def load_model():
-    model_name = "Qwen/Qwen2.5-Math-1.5B-Instruct"
-    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto")
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
-    return model, tokenizer
-
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
-    
-    for user_msg, bot_reply in history:
-        messages.append({"role": "user", "content": user_msg})
-        if bot_reply:
-            messages.append({"role": "assistant", "content": bot_reply})
-    
-    messages.append({"role": "user", "content": message})
-    
-    text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
-    model_inputs = tokenizer([text], return_tensors="pt").to("cuda")
-    
-    generated_ids = model.generate(
-        **model_inputs,
-        # max_new_tokens=max_tokens,
-        # temperature=temperature,
-        # top_p=top_p,
-    )
-    
-    response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-    return response
-
-# Load model and tokenizer
+import tempfile
+import secrets
+from pathlib import Path
+from transformers import AutoModelForCausalLM, AutoTokenizer, BlipForConditionalGeneration, AutoProcessor
+from PIL import Image
+
+# Load Vision-Language Model
+vl_model_name = "Salesforce/blip-image-captioning-large"
+vl_model = BlipForConditionalGeneration.from_pretrained(vl_model_name)
+vl_processor = AutoProcessor.from_pretrained(vl_model_name)
+
+# Load Text Model
+model_name = "Qwen/Qwen2.5-Math-1.5B-Instruct"
 device = "cuda" if torch.cuda.is_available() else "cpu"
-model, tokenizer = load_model()
+model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float16, device_map="auto")
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+math_messages = []
+
+def process_image(image, shouldConvert=False):
+    global math_messages
+    math_messages = []  # Reset when uploading an image
+
+    if shouldConvert:
+        new_img = Image.new('RGB', size=(image.width, image.height), color=(255, 255, 255))
+        new_img.paste(image, (0, 0), mask=image)
+        image = new_img
+
+    # Convert the image to tensor
+    inputs = vl_processor(images=image, return_tensors="pt")
+    output = vl_model.generate(**inputs)
+    description = vl_processor.batch_decode(output, skip_special_tokens=True)[0]
+
+    return f"Math-related content detected: {description}"
+
+def get_math_response(image_description, user_question):
+    global math_messages
+    if not math_messages:
+        math_messages.append({'role': 'system', 'content': 'You are a helpful math assistant.'})
+    math_messages = math_messages[:1]
+    content = f'Image description: {image_description}\n\n' if image_description else ''
+    query = f"{content}User question: {user_question}"
+    math_messages.append({'role': 'user', 'content': query})
+    model_inputs = tokenizer(query, return_tensors="pt").to(device)
+    output = model.generate(**model_inputs, max_new_tokens=512)
+    answer = tokenizer.decode(output[0], skip_special_tokens=True)
+    yield answer.replace("\\", "\\\\")
+    math_messages.append({'role': 'assistant', 'content': answer})
+
+def math_chat_bot(image, sketchpad, question, state):
+    current_tab_index = state["tab_index"]
+    image_description = None
+    if current_tab_index == 0:
+        if image is not None:
+            image_description = process_image(image)
+    elif current_tab_index == 1:
+        if sketchpad and sketchpad["composite"]:
+            image_description = process_image(sketchpad["composite"], True)
+    yield from get_math_response(image_description, question)
 
 demo = gr.ChatInterface(
-    respond,
+    math_chat_bot,
     additional_inputs=[
         gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        # gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        # gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        # gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
+        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
     ],
 )