Update main.py

main.py

@@ -1,247 +1,120 @@
-import os  
-import shutil  
-from flask import Flask, render_template, request, jsonify  
-from llama_index.core import StorageContext, load_index_from_storage, VectorStoreIndex, SimpleDirectoryReader, ChatPromptTemplate, Settings  
-from llama_index.llms.huggingface import HuggingFaceInferenceAPI  
-from llama_index.embeddings.huggingface import HuggingFaceEmbedding  
-from huggingface_hub import InferenceClient  
-from transformers import AutoTokenizer, AutoModel
+import os
+from flask import Flask, render_template, request, jsonify, send_file
+from transformers import AutoModelForCausalLM, AutoTokenizer
 from deep_translator import GoogleTranslator
-
-
-# Ensure HF_TOKEN is set  
-HF_TOKEN = os.getenv("HF_TOKEN")  
-if not HF_TOKEN:  
-    raise ValueError("HF_TOKEN environment variable not set.")  
-
-repo_id = "meta-llama/Meta-Llama-3-8B-Instruct"  
-llm_client = InferenceClient(  
-    model=repo_id,  
-    token=HF_TOKEN,  
-)  
-
-# Configure Llama index settings  
-Settings.llm = HuggingFaceInferenceAPI(  
-    model_name=repo_id,  
-    tokenizer_name=repo_id,  
-    context_window=3000,  
-    token=HF_TOKEN,  
-    max_new_tokens=512,  
-    generate_kwargs={"temperature": 0.1},  
-)  
-# Settings.embed_model = HuggingFaceEmbedding(  
-#     model_name="BAAI/bge-small-en-v1.5"  
-# )  
-# Replace the embedding model with XLM-R
-# Settings.embed_model = HuggingFaceEmbedding(
-#     model_name="xlm-roberta-base"  # XLM-RoBERTa model for multilingual support
-# )
-Settings.embed_model = HuggingFaceEmbedding(
-    model_name="sentence-transformers/paraphrase-multilingual-mpnet-base-v2"
-)
-
-# Configure tokenizer and model if required
-tokenizer = AutoTokenizer.from_pretrained("xlm-roberta-base")
-model = AutoModel.from_pretrained("xlm-roberta-base")
-
-PERSIST_DIR = "db"  
-PDF_DIRECTORY = 'data'  
-
-# Ensure directories exist  
-os.makedirs(PDF_DIRECTORY, exist_ok=True)  
-os.makedirs(PERSIST_DIR, exist_ok=True)  
-chat_history = []  
-current_chat_history = []  
-
-def data_ingestion_from_directory():  
-    # Clear previous data by removing the persist directory  
-    if os.path.exists(PERSIST_DIR):  
-        shutil.rmtree(PERSIST_DIR)  # Remove the persist directory and all its contents  
-    
-    # Recreate the persist directory after removal  
-    os.makedirs(PERSIST_DIR, exist_ok=True)  
-    
-    # Load new documents from the directory  
-    new_documents = SimpleDirectoryReader(PDF_DIRECTORY).load_data()  
-    
-    # Create a new index with the new documents  
-    index = VectorStoreIndex.from_documents(new_documents)  
-    
-    # Persist the new index  
-    index.storage_context.persist(persist_dir=PERSIST_DIR)  
-
-# def handle_query(query):  
-#     context_str = ""  
-    
-#     # Build context from current chat history  
-#     for past_query, response in reversed(current_chat_history):  
-#         if past_query.strip():  
-#             context_str += f"User asked: '{past_query}'\nBot answered: '{response}'\n"  
-    
-#     chat_text_qa_msgs = [
-#         (
-#             "user",
-#             """
-#             You are the Taj Hotel voice chatbot and your name is Taj hotel helper. Your goal is to provide accurate, professional, and helpful answers to user queries based on the Taj hotel data. Always ensure your responses are clear and concise. Give response within 10-15 words only. You need to give an answer in the same language used by the user.       
-#             {context_str}
-#             Question:
-#             {query_str}
-#             """
-#         )
-#     ]
-
-
-    
-#     text_qa_template = ChatPromptTemplate.from_messages(chat_text_qa_msgs)  
-    
-#     storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR)  
-#     index = load_index_from_storage(storage_context)  
-#     # context_str = ""  
-    
-#     # # Build context from current chat history  
-#     # for past_query, response in reversed(current_chat_history):  
-#     #     if past_query.strip():  
-#     #         context_str += f"User asked: '{past_query}'\nBot answered: '{response}'\n"  
-
-#     query_engine = index.as_query_engine(text_qa_template=text_qa_template, context_str=context_str)  
-#     print(f"Querying: {query}")  
-#     answer = query_engine.query(query)  
-
-#     # Extracting the response  
-#     if hasattr(answer, 'response'):  
-#         response = answer.response  
-#     elif isinstance(answer, dict) and 'response' in answer:  
-#         response = answer['response']  
-#     else:  
-#         response = "I'm sorry, I couldn't find an answer to that."  
-
-#     # Append to chat history  
-#     current_chat_history.append((query, response))  
-#     return response
+from gtts import gTTS
+import uuid
+
+# Initialize Flask app
+app = Flask(__name__)
+
+# Load the DeepSeek-V3 model and tokenizer
+model_name = "deepseek-ai/DeepSeek-V3"
+tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
+
+# Supported languages for translation
+SUPPORTED_LANGUAGES = {
+    "english": "en", "hindi": "hi", "bengali": "bn", "telugu": "te", "marathi": "mr",
+    "tamil": "ta", "gujarati": "gu", "kannada": "kn", "malayalam": "ml", "punjabi": "pa",
+    "odia": "or", "urdu": "ur", "assamese": "as", "sanskrit": "sa", "arabic": "ar",
+    "chinese": "zh-CN", "dutch": "nl", "french": "fr", "filipino": "tl", "greek": "el",
+    "indonesian": "id", "italian": "it", "japanese": "ja", "korean": "ko", "latin": "la",
+    "nepali": "ne", "portuguese": "pt", "romanian": "ro", "russian": "ru", "spanish": "es",
+    "swedish": "sv", "thai": "th", "ukrainian": "uk", "turkish": "tr"
+}
+
+# Temporary directory for audio files
+AUDIO_DIR = "audio"
+os.makedirs(AUDIO_DIR, exist_ok=True)
+
+# Handle user queries using DeepSeek-V3
 def handle_query(query):
-    chat_text_qa_msgs = [
-        (
-            "user",
-            """
-            You are the Hotel voice chatbot and your name is hotel helper. Your goal is to provide accurate, professional, and helpful answers to user queries based on the hotel's data. Always ensure your responses are clear and concise. Give response within 10-15 words only. You need to give an answer in the same language used by the user.       
-            {context_str}
-            Question:
-            {query_str}
-            """
-        )
-    ]
-    text_qa_template = ChatPromptTemplate.from_messages(chat_text_qa_msgs)
-    
-    storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR)
-    index = load_index_from_storage(storage_context)
-    context_str = ""
-    for past_query, response in reversed(current_chat_history):
-        if past_query.strip():
-            context_str += f"User asked: '{past_query}'\nBot answered: '{response}'\n"
-
-    query_engine = index.as_query_engine(text_qa_template=text_qa_template, context_str=context_str)
-    print(query)
-    answer = query_engine.query(query)
-
-    if hasattr(answer, 'response'):
-        response = answer.response
-    elif isinstance(answer, dict) and 'response' in answer:
-        response = answer['response']
-    else:
-        response = "Sorry, I couldn't find an answer."
-    current_chat_history.append((query, response))
-    return response
-
-app = Flask(__name__)  
-
-# Data ingestion  
-data_ingestion_from_directory()  
-
-# Generate Response
-def generate_response(query, language):  
-    try:  
-        # Call the handle_query function to get the response  
+    # Tokenize the input query
+    inputs = tokenizer(query, return_tensors="pt")
+
+    # Generate a response using the DeepSeek-V3 model
+    outputs = model.generate(
+        inputs.input_ids,
+        max_length=50,  # Adjust the max length as needed
+        num_return_sequences=1,
+        temperature=0.7,  # Adjust temperature for creativity
+        pad_token_id=tokenizer.eos_token_id  # Ensure proper padding
+    )
+
+    # Decode the generated response
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    return response.strip()
+
+# Generate response and translate if needed
+def generate_response(query, language):
+    try:
+        # Get response from DeepSeek-V3
         bot_response = handle_query(query)
-        
-        # Map of supported languages
-        supported_languages = {
-            "hindi": "hi",
-            "bengali": "bn",
-            "telugu": "te",
-            "marathi": "mr",
-            "tamil": "ta",
-            "gujarati": "gu",
-            "kannada": "kn",
-            "malayalam": "ml",
-            "punjabi": "pa",
-            "odia": "or",
-            "urdu": "ur",
-            "assamese": "as",
-            "sanskrit": "sa",
-            "arabic": "ar",
-            "australian": "en-AU",
-            "bangla-india": "bn-IN",
-            "chinese": "zh-CN",
-            "dutch": "nl",
-            "french": "fr",
-            "filipino": "tl",
-            "greek": "el",
-            "indonesian": "id",
-            "italian": "it",
-            "japanese": "ja",
-            "korean": "ko",
-            "latin": "la",
-            "nepali": "ne",
-            "portuguese": "pt",
-            "romanian": "ro",
-            "russian": "ru",
-            "spanish": "es",
-            "swedish": "sv",
-            "thai": "th",
-            "ukrainian": "uk",
-            "turkish": "tr"
-        }
-        
-        # Initialize the translated text
-        translated_text = bot_response
-        
-        # Translate only if the language is supported and not English
-        try:
-            if language in supported_languages:
-                target_lang = supported_languages[language]
-                translated_text = GoogleTranslator(source='en', target=target_lang).translate(bot_response)
-                print(translated_text)
-            else:
-                print(f"Unsupported language: {language}")
-        except Exception as e:
-            # Handle translation errors
-            print(f"Translation error: {e}")
-            translated_text = "Sorry, I couldn't translate the response."
-        
-        # Append to chat history
-        chat_history.append((query, translated_text))
-        return translated_text  
-    except Exception as e:  
+
+        # Translate response if the language is not English
+        target_lang = SUPPORTED_LANGUAGES.get(language.lower(), "en")
+        if target_lang != "en":
+            try:
+                bot_response = GoogleTranslator(source='en', target=target_lang).translate(bot_response)
+            except Exception as e:
+                print(f"Translation error: {e}")
+                bot_response = "Sorry, I couldn't translate the response."
+
+        return bot_response
+    except Exception as e:
         return f"Error fetching the response: {str(e)}"
 
-# Route for the homepage  
-@app.route('/')  
-def index():  
-    return render_template('index.html')  
-
-# Route to handle chatbot messages  
-@app.route('/chat', methods=['POST'])  
-def chat():  
-    try:  
-        user_message = request.json.get("message")
-        language = request.json.get("language")
-        if not user_message:  
-            return jsonify({"response": "Please say something!"})  
-
-        bot_response = generate_response(user_message,language)  
-        return jsonify({"response": bot_response})  
-    except Exception as e:  
-        return jsonify({"response": f"An error occurred: {str(e)}"})  
-
-if __name__ == '__main__':  
+# Convert text to speech using gTTS
+def text_to_speech(text, lang="en"):
+    try:
+        # Generate a unique filename for the audio file
+        audio_filename = f"{uuid.uuid4()}.mp3"
+        audio_path = os.path.join(AUDIO_DIR, audio_filename)
+
+        # Create gTTS object and save the audio file
+        tts = gTTS(text=text, lang=lang)
+        tts.save(audio_path)
+
+        return audio_path
+    except Exception as e:
+        print(f"Error generating speech: {e}")
+        return None
+
+# Flask routes
+@app.route('/')
+def index():
+    return render_template('index.html')
+
+@app.route('/chat', methods=['POST'])
+def chat():
+    try:
+        data = request.json
+        user_message = data.get("message")
+        language = data.get("language", "english").lower()
+
+        if not user_message:
+            return jsonify({"response": "Please say something!", "audio_url": None})
+
+        # Generate response
+        bot_response = generate_response(user_message, language)
+
+        # Convert response to speech
+        target_lang = SUPPORTED_LANGUAGES.get(language, "en")
+        audio_path = text_to_speech(bot_response, lang=target_lang)
+
+        if audio_path:
+            audio_url = f"/audio/{os.path.basename(audio_path)}"
+        else:
+            audio_url = None
+
+        return jsonify({"response": bot_response, "audio_url": audio_url})
+    except Exception as e:
+        return jsonify({"response": f"An error occurred: {str(e)}", "audio_url": None})
+
+@app.route('/audio/<filename>')
+def serve_audio(filename):
+    return send_file(os.path.join(AUDIO_DIR, filename), mimetype="audio/mp3")
+
+if __name__ == '__main__':
     app.run(debug=True)