Update app.py

app.py

@@ -1,61 +1,81 @@
 import os
-import streamlit as st
 from huggingface_hub import login, hf_hub_download
-from datasets import load_dataset
+import pandas as pd
+import gradio as gr
 from llama_cpp import Llama
 import chromadb
-from chromadb.config import Settings  # Added import for Settings
 from sentence_transformers import SentenceTransformer
+from deep_translator import GoogleTranslator  # Changed from googletrans to deep_translator
+import re
+import requests  # Import the requests library
 
-# Load Hugging Face token from environment variable
-hf_token = os.getenv("HF_TOKEN")
-if hf_token:
-    login(token=hf_token)
-else:
-    raise ValueError("HF_TOKEN is not set. Please add it to your Hugging Face Space secrets.")
-
-# Load dataset
-dataset = load_dataset("Maryem2025/final_dataset")
 
-# Initialize Llama model
-llm = Llama(
-    model_path=hf_hub_download(
+# Charger le token depuis les secrets
+hf_token = os.getenv("HF_TOKEN")
+login(token=hf_token)
+
+# Charger le dataset depuis un fichier CSV local
+csv_file = "/content/indian_food (1).csv"
+try:
+    df = pd.read_csv(csv_file)
+    print("Dataset chargé avec succès depuis le fichier CSV local.")
+except FileNotFoundError:
+    print(f"Erreur: Fichier CSV non trouvé à l'emplacement: {csv_file}")
+    exit()
+except Exception as e:
+    print(f"Erreur lors du chargement du CSV: {e}")
+    exit()
+
+# Initialisation du modèle Llama
+llm = None  # Initialize to None
+try:
+    # Use /tmp for the model path within Hugging Face Spaces
+    model_path = hf_hub_download(
         repo_id="TheBloke/CapybaraHermes-2.5-Mistral-7B-GGUF",
         filename="capybarahermes-2.5-mistral-7b.Q2_K.gguf",
-        token=hf_token  # Ensure the token is passed for authentication
-    ),
-    n_ctx=2048,
-)
+        cache_dir="/tmp"  # Store the model in /tmp
+    )
+
+    llm = Llama(
+        model_path=model_path,
+        n_ctx=2048,
+    )
+    print("Llama model loaded successfully.")
+except Exception as e:
+    print(f"Error loading Llama model: {e}")
 
-# Initialize ChromaDB
+# Initialisation de ChromaDB Vector Store
 class VectorStore:
     def __init__(self, collection_name):
         self.embedding_model = SentenceTransformer('sentence-transformers/multi-qa-MiniLM-L6-cos-v1')
-        # Use Settings to configure persistence
-        self.chroma_client = chromadb.Client(Settings(
-            chroma_db_impl="duckdb+parquet",
-            persist_directory="./chroma_db"  # Ensure persistence
-        ))
-        if collection_name in [c.name for c in self.chroma_client.list_collections()]:
-            self.chroma_client.delete_collection(name=collection_name)
+        self.chroma_client = chromadb.Client()
+
+        if collection_name in self.chroma_client.list_collections():
+            self.chroma_client.delete_collection(collection_name)
         self.collection = self.chroma_client.create_collection(name=collection_name)
 
-    def populate_vectors(self, dataset):
-        titles = dataset['train']['title'][:2000]
-        servings = dataset['train']['servings'][:2000]
-        total_times = dataset['train']['total_time'][:2000]
-        courses = dataset['train']['course'][:2000]
-        sections = dataset['train']['sections'][:2000]
-        instructions = dataset['train']['instructions'][:2000]
-        cuisines = dataset['train']['cuisine'][:2000]
-        calories = dataset['train']['calories'][:2000]
+    def populate_vectors(self, df):
+        titles = df['name'].tolist()
+        ingredients = df['ingredients'].tolist()
+        diets = df['diet'].tolist()
+        prep_times = df['prep_time'].tolist()
+
+        # Load nutritional information, handling potentially missing columns and types
+        calories = df['calories'].astype(str).tolist() if 'calories' in df else ['None'] * len(df)
+        sugar = df['sugar'].astype(str).tolist() if 'sugar' in df else ['None'] * len(df)
+        gluten = df['gluten'].astype(str).tolist() if 'gluten' in df else ['None'] * len(df)
+
+        titles = titles[:2000]
+        ingredients = ingredients[:2000]
+        diets = diets[:2000]
+        prep_times = prep_times[:2000]
+        calories = calories[:2000]
+        sugar = sugar[:2000]
+        gluten = gluten[:2000]
 
         texts = [
-            f"Title: {title}. Servings: {serving}. Total Time: {total_time} minutes. "
-            f"Course: {course}. Sections: {section}. Instructions: {instruction}. "
-            f"Cuisine: {cuisine}. Calories: {calorie}."
-            for title, serving, total_time, course, section, instruction, cuisine, calorie 
-            in zip(titles, servings, total_times, courses, sections, instructions, cuisines, calories)
+            f"Recipe: {title}. Ingredients: {ingredient}. Diet: {diet}. Prep Time: {prep_time} minutes. Calories: {calorie}. Sugar: {sugar}. Gluten: {gluten}."
+            for title, ingredient, diet, prep_time, calorie, sugar, gluten in zip(titles, ingredients, diets, prep_times, calories, sugar, gluten)
         ]
 
         for i, item in enumerate(texts):
@@ -65,48 +85,138 @@ class VectorStore:
     def search_context(self, query, n_results=1):
         query_embedding = self.embedding_model.encode([query]).tolist()
         results = self.collection.query(query_embeddings=query_embedding, n_results=n_results)
-        return results['documents'][0]  # Adjusted to access the correct document
-
-# Initialize and populate vector store
-vector_store = VectorStore("embedding_vector")
-vector_store.populate_vectors(dataset)
-
-# Define function for generating text
-def generate_text(message):
-    context_results = vector_store.search_context(message, n_results=1)
-    context = context_results[0] if context_results else ""
+        return results['documents']
+
+# Initialisation du store de vecteurs et peuplement
+vector_store = None  # Initialize to None
+try:
+    vector_store = VectorStore("indian_food_embedding")
+    vector_store.populate_vectors(df)
+    print("Vector store initialized and populated.")
+except Exception as e:
+    print(f"Error initializing or populating vector store: {e}")
+
+
+# Replace the translate_text function with this new version
+def translate_text(text, target_language='en'):
+    """Translates the given text to the target language."""
+    try:
+        if target_language == 'en':
+            translator = GoogleTranslator(source='auto', target='en')
+        else:
+            translator = GoogleTranslator(source='en', target=target_language)
+
+        translated_text = translator.translate(text)
+        return translated_text
+    except Exception as e:
+        print(f"Translation error: {e}")
+        print(f"Detailed error: {type(e).__name__}, {e}")  # Print more details for debugging.
+        return text  # Return original text if translation fails
+
+def generate_text(message, max_tokens=600, temperature=0.3, top_p=0.95,
+                  gluten_free=False, dairy_free=False, allergies="", input_language='en'):  # Added input_language
+
+    if llm is None:
+        return "Error: Llama model could not be loaded. Please check the console for errors."
+
+    if vector_store is None:
+        return "Error: Vector store could not be initialized. Please check the console for errors."
+
+    # Translate the input message to English
+    message_en = message
+    if input_language != 'en':
+      try:
+        message_en = translate_text(message, target_language='en')
+      except Exception as e:
+        print(f"Error translating input message: {e}")
+        return "Error translating input.  Please try again in English."
+
+
+
+    context = ""
+    query = message_en
+    if gluten_free:
+        query += " gluten-free"
+    if dairy_free:
+        query += " dairy-free"
+    if allergies:
+        query += f" avoid ingredients: {allergies}"
+
+    try:
+        context_results = vector_store.search_context(query, n_results=1)
+        if context_results and isinstance(context_results, list):
+            context = context_results[0] if context_results else ""
+        else:
+            context = ""  # or handle the error appropriately
+            print("Warning: No context found or invalid context format.")
+    except Exception as e:
+        return f"Error searching vector store: {e}"
 
     prompt_template = (
-        f"SYSTEM: You are a recipe generating bot.\n"
-        f"SYSTEM: {context}\n"
-        f"USER: {message}\n"
+        f"SYSTEM: You are a helpful recipe generating bot specializing in Indian cuisine, assisting with dietary restrictions.\n"
+        f"SYSTEM: Here is some context:\n{context}\n"
+        f"USER: {message_en}\n"  # Use the English translated message
         f"ASSISTANT:\n"
     )
 
-    output = llm(
-        prompt_template,
-        temperature=0.3,
-        top_p=0.95,
-        top_k=40,
-        repeat_penalty=1.1,
-        max_tokens=600,
-    )
-
-    input_string = output['choices'][0]['text'].strip()
-    cleaned_text = input_string.strip("[]'").replace('\\n', '\n')
-    return cleaned_text
-
-# Streamlit UI
-st.title("JOSHI’s AI Chef 🍽️")
-st.write("Generate recipes using AI powered by Hugging Face and ChromaDB!")
-
-user_input = st.text_area("Enter ingredients or ask for a recipe:", "")
+    try:
+        output = llm(
+            prompt_template,
+            temperature=temperature,
+            top_p=top_p,
+            top_k=40,
+            repeat_penalty=1.1,
+            max_tokens=max_tokens,
+        )
+
+        input_string = output['choices'][0]['text'].strip()
+        cleaned_text = input_string.strip("[]'").replace('\\n', '\n')
+        continuous_text = '\n'.join(cleaned_text.split('\n'))
+
+        # Translate the output back to the input language
+        output_text = continuous_text
+        if input_language != 'en':
+            try:
+                output_text = translate_text(continuous_text, target_language=input_language)
+            except Exception as e:
+                print(f"Error translating output message: {e}")
+                output_text = "Error translating output.  Here is the English version:\n\n" + continuous_text
+
+
+        # Gluten Check on Output
+        if context and isinstance(context, str):
+            context_lower = context.lower()
+            if "gluten: yes" in context_lower:
+                output_text += "\n\nWarning: This recipe contains gluten."
+            elif "gluten: no" in context_lower:
+                output_text += "\n\nGood news! This recipe is gluten-free."
+
+        return output_text
+
+    except Exception as e:
+        return f"Error generating text: {e}"
+
+demo = gr.Interface(
+    fn=generate_text,
+    inputs=[
+        gr.Textbox(lines=2, placeholder="Enter your message here...", label="Message"),
+        gr.Slider(minimum=50, maximum=1000, value=600, step=50, label="Max Tokens"),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.3, step=0.1, label="Temperature"),
+        gr.Slider(minimum=0.7, maximum=1.0, value=0.95, step=0.05, label="Top P"),
+        gr.Checkbox(label="Gluten-Free"),
+        gr.Checkbox(label="Dairy-Free"),
+        gr.Textbox(lines=1, placeholder="e.g., peanuts, shellfish", label="Allergies (comma-separated)"),
+        gr.Dropdown(choices=['en', 'hi'], value='en', label="Input Language (en=English, hi=Hindi/Hinglish)"), # Added language selection
+    ],
+    outputs=gr.Textbox(label="Generated Text"),
+    title="Indian Recipe Bot",
+    description="Running LLM with context retrieval from ChromaDB.  Supports dietary restrictions, allergies, and Hinglish input/output!",
+    examples=[
+        ["mujhe chawal aur dal hai, main kya bana sakta hoon jo gluten-free ho?", 600, 0.3, 0.95, True, False, "", 'hi'],
+        ["Suggest a vegetarian dish with spinach and no nuts.", 600, 0.3, 0.95, False, False, "nuts", 'en'],
+    ],
+    cache_examples=False,
+)
 
-if st.button("Generate Recipe"):
-    if user_input:
-        with st.spinner("Generating recipe... 🍲"):
-            response = generate_text(user_input)
-        st.subheader("Generated Recipe:")
-        st.write(response)
-    else:
-        st.warning("Please enter a message.")
+if __name__ == "__main__":
+    demo.launch()