Update app.py

app.py

@@ -82,20 +82,104 @@ client_gemma = InferenceClient("google/gemma-1.1-7b-it")
 client_mixtral = InferenceClient("NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO")
 client_llama = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct")
 
-# Define the main chat function
+import gradio as gr
+from huggingface_hub import InferenceClient
+import json
+import uuid
+from PIL import Image
+from bs4 import BeautifulSoup
+import requests
+import random
+from transformers import LlavaProcessor, LlavaForConditionalGeneration, TextIteratorStreamer
+from threading import Thread
+import re
+import time 
+import torch
+import cv2
+
+model_id = "llava-hf/llava-interleave-qwen-0.5b-hf"
+
+processor = LlavaProcessor.from_pretrained(model_id)
+
+model = LlavaForConditionalGeneration.from_pretrained(model_id)
+model.to("cpu")
+
+
+def llava(message, history):
+    if message["files"]:
+        image = message["files"][0]    
+    else:
+        for hist in history:
+            if isinstance(hist[0], tuple):
+                image = hist[0][0]
+        
+    txt = message["text"]
+        
+    gr.Info("Analyzing image")
+    image = Image.open(image).convert("RGB")
+    prompt = f"<|im_start|>user <image>\n{txt}<|im_end|><|im_start|>assistant"
+    
+    inputs = processor(prompt, image, return_tensors="pt")
+    
+    # Return the dictionary format expected by MultimodalTextbox
+    return {"text": txt, "files": [image]}
+
+
+def extract_text_from_webpage(html_content):
+    soup = BeautifulSoup(html_content, 'html.parser')
+    for tag in soup(["script", "style", "header", "footer"]):
+        tag.extract()
+    return soup.get_text(strip=True)
+
+def search(query):
+    term = query
+    start = 0
+    all_results = []
+    max_chars_per_page = 8000
+    with requests.Session() as session:
+        resp = session.get(
+            url="https://www.google.com/search",
+            headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0"},
+            params={"q": term, "num": 3, "udm": 14},
+            timeout=5,
+            verify=None,
+        )
+        resp.raise_for_status()
+        soup = BeautifulSoup(resp.text, "html.parser")
+        result_block = soup.find_all("div", attrs={"class": "g"})
+        for result in result_block:
+            link = result.find("a", href=True)
+            link = link["href"]
+            try:
+                webpage = session.get(link, headers={"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0"}, timeout=5, verify=False)
+                webpage.raise_for_status()
+                visible_text = extract_text_from_webpage(webpage.text)
+                if len(visible_text) > max_chars_per_page:
+                    visible_text = visible_text[:max_chars_per_page]
+                all_results.append({"link": link, "text": visible_text})
+            except requests.exceptions.RequestException:
+                all_results.append({"link": link, "text": None})
+    return all_results
+
+# Initialize inference clients for different models
+client_gemma = InferenceClient("google/gemma-1.1-7b-it")
+client_mixtral = InferenceClient("NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO")
+client_llama = InferenceClient("meta-llama/Meta-Llama-3-8B-Instruct")
+
+# Corrected the response format to ensure it works with Gradio's multimodal interface.
 def respond(message, history):
     func_caller = []
 
     user_prompt = message
     # Handle image processing
-    if message["files"]:
+    if message.get("files"):
         inputs = llava(message, history)
         streamer = TextIteratorStreamer(processor, skip_prompt=True, **{"skip_special_tokens": True})
         generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)
 
         thread = Thread(target=model.generate, kwargs=generation_kwargs)
         thread.start()
-    
+
         buffer = ""
         for new_text in streamer:
             buffer += new_text
@@ -109,15 +193,14 @@ def respond(message, history):
         ]
     
         message_text = message["text"]
-        func_caller.append({"role": "user", "content": f'[SYSTEM]You are a helpful AI assistant for a discord server called Stars Kingdom, your job is to have fun help users and listen to what they say or wan\'t you to act. You have been created by the discord server owner named Star. You have access to the following functions: \n {str(functions_metadata)}\n\nTo use these functions respond with:\n<functioncall> {{ "name": "function_name", "arguments": {{ "arg_1": "value_1", "arg_1": "value_1", ... }} }}  </functioncall>  [USER] {message_text}'})
+        func_caller.append({"role": "user", "content": f'[SYSTEM]You are a helpful AI assistant for a discord server called Stars Kingdom, your job is to have fun help users and listen to what they say or want you to act. You have been created by the discord server owner named Star. You have access to the following functions: \n {str(functions_metadata)}\n\nTo use these functions respond with:\n<functioncall> {{ "name": "function_name", "arguments": {{ "arg_1": "value_1", "arg_2": "value_2", ... }} }}  </functioncall>  [USER] {message_text}'})
 
-    
         response = client_gemma.chat_completion(func_caller, max_tokens=150)
         response = str(response)
         try:
             response = response[int(response.find("{")):int(response.index("</"))]
         except:
-            print("A error occured")
+            print("An error occurred")
         response = response.replace("\\n", "")
         response = response.replace("\\'", "'")
         response = response.replace('\\"', '"')
@@ -133,11 +216,11 @@ def respond(message, history):
                 web_results = search(query)
                 gr.Info("Extracting relevant Info")
                 web2 = ' '.join([f"Link: {res['link']}\nText: {res['text']}\n\n" for res in web_results])
-                messages = f"<|im_start|>system\nYou are a helpful assistant made by Star. You are provided with WEB results from which you can find informations to answer users query in Structured and More better way. You do not say Unnecesarry things Only say thing which is important and relevant. You also Expert in fun and helping people and also learn and try to answer from contexts related to previous question. Try your best to give best response possible to user. You also try to show emotions using Emojis and reply like human but only does this rarely don't overuse this, use short forms, friendly tone and emotions.<|im_end|>"
+                messages = f"<|im_start|>system\nYou are a helpful assistant made by Star. You are provided with WEB results from which you can find informations to answer users query in a structured and better way. Only respond with what’s important!<|im_end|>"
                 for msg in history:
                     messages += f"\n<|im_start|>user\n{str(msg[0])}<|im_end|>"
                     messages += f"\n<|im_start|>assistant\n{str(msg[1])}<|im_end|>"
-                messages+=f"\n<|im_start|>user\n{message_text}<|im_end|>\n<|im_start|>web_result\n{web2}<|im_end|>\n<|im_start|>assistant\n"
+                messages += f"\n<|im_start|>user\n{message_text}<|im_end|>\n<|im_start|>web_result\n{web2}<|im_end|>\n<|im_start|>assistant\n"
                 stream = client_mixtral.text_generation(messages, max_new_tokens=2000, do_sample=True, stream=True, details=True, return_full_text=False)
                 output = ""
                 for response in stream:
@@ -161,17 +244,18 @@ def respond(message, history):
 
                 thread = Thread(target=model.generate, kwargs=generation_kwargs)
                 thread.start()
-    
+
                 buffer = ""
                 for new_text in streamer:
                     buffer += new_text
                     yield buffer
             else:
-                messages = f"<|start_header_id|>system\nYou are a helpful assistant made by Star. You are provided with WEB results from which you can find informations to answer users query in Structured and More better way. You do not say Unnecesarry things Only say thing which is important and relevant. You also Expert in fun and helping people and also learn and try to answer from contexts related to previous question. Try your best to give best response possible to user. You also try to show emotions using Emojis and reply like human but only does this rarely don't overuse this, use short forms, friendly tone and emotions.<|end_header_id|>"
+                # Default response from llama model if no specific function matched
+                messages = f"<|im_start|>system\nYou are a helpful assistant made by Star.<|im_end|>"
                 for msg in history:
-                    messages += f"\n<|start_header_id|>user\n{str(msg[0])}<|end_header_id|>"
-                    messages += f"\n<|start_header_id|>assistant\n{str(msg[1])}<|end_header_id|>"
-                messages+=f"\n<|start_header_id|>user\n{message_text}<|end_header_id|>\n<|start_header_id|>assistant\n"
+                    messages += f"\n<|im_start|>user\n{str(msg[0])}<|im_end|>"
+                    messages += f"\n<|im_start|>assistant\n{str(msg[1])}<|im_end|>"
+                messages += f"\n<|im_start|>user\n{message_text}<|im_end|>\n<|im_start|>assistant\n"
                 stream = client_llama.text_generation(messages, max_new_tokens=2000, do_sample=True, stream=True, details=True, return_full_text=False)
                 output = ""
                 for response in stream:
@@ -179,11 +263,11 @@ def respond(message, history):
                         output += response.token.text
                         yield output
         except:
-            messages = f"<|start_header_id|>system\nYou are a helpful assistant made by Star. You answers users query like human friend. You also Expert in fun and helping people and also learn and try to answer from contexts related to previous question. Try your best to give best response possible to user. You also try to show emotions using Emojis and reply like human, use short forms, friendly tone and emotions.<|end_header_id|>"
+            messages = f"<|im_start|>system\nYou are a helpful assistant made by Star. You answer users' queries like a human friend.<|im_end|>"
             for msg in history:
-                messages += f"\n<|start_header_id|>user\n{str(msg[0])}<|end_header_id|>"
-                messages += f"\n<|start_header_id|>assistant\n{str(msg[1])}<|end_header_id|>"
-            messages+=f"\n<|start_header_id|>user\n{message_text}<|end_header_id|>\n<|start_header_id|>assistant\n"
+                messages += f"\n<|im_start|>user\n{str(msg[0])}<|im_end|>"
+                messages += f"\n<|im_start|>assistant\n{str(msg[1])}<|im_end|>"
+            messages += f"\n<|im_start|>user\n{message_text}<|im_end|>\n<|im_start|>assistant\n"
             stream = client_llama.text_generation(messages, max_new_tokens=2000, do_sample=True, stream=True, details=True, return_full_text=False)
             output = ""
             for response in stream:
@@ -191,11 +275,11 @@ def respond(message, history):
                     output += response.token.text
                     yield output
 
-# Create the Gradio interface
+# Gradio Chat Interface
 demo = gr.ChatInterface(
     fn=respond,
     chatbot=gr.Chatbot(show_copy_button=True, likeable=True, layout="panel"),
-    description ="# Star's Kingdom ai bot",
+    description="AI assistant for Stars Kingdom",
     textbox=gr.MultimodalTextbox(),
     multimodal=True,
     concurrency_limit=200,
@@ -204,8 +288,9 @@ demo = gr.ChatInterface(
         {"text": "What's the preferred shirt color for an interview?",},
         {"text": "How can I dress more smartly?",},
         {"text": "Tell about some good accessories for a traditional Indian wedding",},
-        {"text": "What's the colour of the frock in the given image?", "files": ["./frock.png"]},
+        {"text": "What's the color of the frock in the given image?", "files": ["./frock.png"]},
     ],
     cache_examples=False,
 )
+
 demo.launch(share=True)