import os os.system('pip install curl_cffi tqdm bitsandbytes tiktoken g4f pinecone-client pandas datasets sentence-transformers') # Setup and load your keys import os from g4f import ChatCompletion #from google.colab import userdata from pinecone import Pinecone import pandas as pd from datasets import Dataset from sentence_transformers import SentenceTransformer import gradio as gr model_name = "BAAI/bge-m3" # APIs personales #PINECONE_ENVIRONMENT = us-east-1 #PINECONE_API_KEY = 3a3e9022-381d-436e-84cb-ba93464d283e os.environ["PINECONE_ENVIRONMENT"] = "us-east-1" os.environ["PINECONE_API_KEY"] = "3a3e9022-381d-436e-84cb-ba93464d283e" # Retrieve the Pinecone API key from the user PINECONE_API_KEY = "3a3e9022-381d-436e-84cb-ba93464d283e" # Use the key you set in the secrets PINECONE_ENVIRONMENT = "us-east-1" # Use the environment you set in the secrets # Initialize Pinecone with the API key pc = Pinecone(api_key=PINECONE_API_KEY) # Global variables to store the selected model and dimensions EMBED_MODEL = 'BGE_M3-1024' DIMENSIONS = 1024 # Confirm selection automatically print(f"Model selected: {EMBED_MODEL}") print(f"Dimensions set as: {DIMENSIONS}") # Function to print current selection (can be used in other cells) def print_current_selection(): print(f"Currently selected model: {EMBED_MODEL}") print(f"Dimensions: {DIMENSIONS}") # Establecer el nombre del índice automáticamente INDEX_NAME = 'vestidos' # Obtener la clave API de Pinecone #PINECONE_API_KEY = userdata.get('PINECONE_API_KEY') def connect_to_pinecone(index_name): global INDEX_NAME try: pc = Pinecone(api_key=PINECONE_API_KEY) index = pc.Index(index_name) # Asegurarse de que la conexión se establezca index_stats = index.describe_index_stats() print(f"Successfully connected to Pinecone index '{index_name}'!") print("Index Stats:", index_stats) # Actualizar la variable global INDEX_NAME INDEX_NAME = index_name print(f"Global INDEX_NAME updated to: {INDEX_NAME}") except Exception as e: print(f"Failed to connect to Pinecone index '{index_name}':", str(e)) # Conectar automáticamente al índice "vestidos" connect_to_pinecone(INDEX_NAME) # Función para imprimir el nombre del índice actual (puede ser usada en otras celdas) def print_current_index(): print(f"Current index name: {INDEX_NAME}") # Verificar si las variables globales necesarias están configuradas if 'INDEX_NAME' not in globals() or INDEX_NAME is None: raise ValueError("INDEX_NAME is not set. Please set the index name first.") if 'EMBED_MODEL' not in globals() or EMBED_MODEL is None: raise ValueError("EMBED_MODEL is not set. Please select an embedding model first.") # Inicializar cliente de Pinecone #PINECONE_API_KEY = userdata.get('PINECONE_API_KEY') pc = Pinecone(api_key=PINECONE_API_KEY) # Inicializar el índice de Pinecone index = pc.Index(INDEX_NAME) # Obtener la dimensión del índice index_stats = index.describe_index_stats() vector_dim = index_stats['dimension'] print(f"Index dimension: {vector_dim}") # Definir manualmente los campos de contexto y enlace CONTEXT_FIELDS = ['Etiqueta', 'Pregunta 1', 'Pregunta 2', 'Pregunta 3', 'Respuesta Combinada'] LINK_FIELDS = ['Etiqueta', 'Respuesta Combinada'] # Imprimir confirmación de campos seleccionados print(f"Context fields set to: {CONTEXT_FIELDS}") print(f"Link fields set to: {LINK_FIELDS}") # Función para obtener las selecciones actuales de campos (puede ser usada en otras celdas) def get_field_selections(): return { "CONTEXT_FIELDS": CONTEXT_FIELDS, "LINK_FIELDS": LINK_FIELDS } ##################################### # Check if required global variables are set if 'EMBED_MODEL' not in globals() or EMBED_MODEL is None: raise ValueError("EMBED_MODEL is not set. Please select an embedding model first.") if 'INDEX_NAME' not in globals() or INDEX_NAME is None: raise ValueError("INDEX_NAME is not set. Please create or select an index first.") if 'CONTEXT_FIELDS' not in globals() or 'LINK_FIELDS' not in globals(): raise ValueError("CONTEXT_FIELDS and LINK_FIELDS are not set. Please run the field selection cell first.") # Initialize the Sentence-Transformer model embedding_model = SentenceTransformer(model_name) # Initialize Pinecone with the API key and connect to the index pinecone_client = Pinecone(api_key=PINECONE_API_KEY) index = pinecone_client.Index(INDEX_NAME) # Constants LIMIT = 3750 def vector_search(query): # Generate embedding using Sentence-Transformer model xq = embedding_model.encode(query) # Perform vector search on Pinecone index res = index.query(vector=xq.tolist(), top_k=3, include_metadata=True) if res['matches']: return [ { 'content': ' '.join(f"{k}: {v}" for k, v in match['metadata'].items() if k in CONTEXT_FIELDS and k != 'Etiqueta'), 'metadata': match['metadata'] } for match in res['matches'] if 'metadata' in match ] return [] def create_prompt(query, contexts): prompt_start = "\n\nContexto:\n" prompt_end = f"\n\nPregunta: {query}\nRespuesta:" current_contexts = "\n\n---\n\n".join([context['content'] for context in contexts]) if len(prompt_start + current_contexts + prompt_end) >= LIMIT: # Truncate contexts if they exceed the limit available_space = LIMIT - len(prompt_start) - len(prompt_end) truncated_contexts = current_contexts[:available_space] return prompt_start + truncated_contexts + prompt_end else: return prompt_start + current_contexts + prompt_end def complete(prompt): return [f"Hola"] def check_image_exists(filepath): return os.path.exists(filepath) def chat_function(message, history): # Perform vector search search_results = vector_search(message) # Create prompt with relevant contexts query_with_contexts = create_prompt(message, search_results) # Generate response response = complete(query_with_contexts) partial_message = response[0].split("\n")[0] # Solo tomar la primera línea de la respuesta # Handle the logic for processing tags and images internally relevant_links = [result['metadata'].get(field) for result in search_results for field in LINK_FIELDS if field in result['metadata']] full_response = partial_message image_url = None tags_detected = [] filtered_links = [] if relevant_links: for link in relevant_links: if any(tag in link for tag in ["lila_61", "lila_63", "lila_62", "lila_64", "fuxia_70", "fuxia_71", "fuxia_72", "fuxia_73", "fuxia_74", "melon_68", "melon_66", "melon_67", "melon_65", "vino_19", "vino_20", "barney_69", "loro_27", "lacre_02", "amarillo_03", "amarillo_04", "azulino_11", "azulino_14", "azulino_12", "azulino_13", "beigs_09", "beigs_10", "beigs_07", "beigs_06", "beigs_08", "beigs_05", "marina_32", "marina_29", "marina_28", "marina_31", "marina_30", "rojo_26", "rojo_23", "rojo_21", "rojo_22", "rojo_25", "rojo_24", "celeste_40", "celeste_38", "celeste_39", "celeste_33", "celeste_35", "celeste_37", "celeste_41", "celeste_42", "celeste_34", "celeste_36", "sirenita_01", "marino_18", "marino_17", "marino_16", "marino_15", "rosa_87", "rosa_86", "rosa_79", "rosa_82", "rosa_83", "rosa_78", "rosa_84", "rosa_85", "rosa_75", "rosa_80", "rosa_81", "rosa_77", "rosa_76", "blanco_55", "blanco_56", "blanco_53", "blanco_52", "blanco_57", "blanco_49", "blanco_51", "blanco_60", "blanco_47", "blanco_44", "blanco_50", "blanco_48", "blanco_59", "blanco_43", "blanco_58", "blanco_46", "blanco_45", "blanco_54"]): tags_detected.append(link) # Save the tag but don't display it else: filtered_links.append(link) # Add the first relevant link under a single "Respuestas relevantes" section if filtered_links: full_response += f".\n\nTe detallamos nuestro contenido a continuación:\n" + filtered_links[0] # Now handle the images based on the detected tags tags_to_images = { "lila_61": "/content/lila_61.jpeg", "lila_63": "/content/lila_63.jpeg", "lila_62": "/content/lila_62.jpeg", "lila_64": "/content/lila_64.jpeg", "fuxia_70": "/content/fuxia_70.jpeg", "fuxia_71": "/content/fuxia_71.jpeg", "fuxia_72": "/content/fuxia_72.jpeg", "fuxia_73": "/content/fuxia_73.jpeg", "fuxia_74": "/content/fuxia_74.jpeg", "melon_68": "/content/melon_68.jpeg", "melon_66": "/content/melon_66.jpeg", "melon_67": "/content/melon_67.jpeg", "melon_65": "/content/melon_65.jpeg", "vino_19": "/content/vino_19.jpeg", "vino_20": "/content/vino_20.jpeg", "barney_69": "/content/barney_69.jpeg", "loro_27": "/content/loro_27.png", "lacre_02": "/content/lacre_02.jpeg", "amarillo_03": "/content/amarillo_03.jpeg", "amarillo_04": "/content/amarillo_04.jpeg", "azulino_11": "/content/azulino_11.jpeg", "azulino_14": "/content/azulino_14.jpeg", "azulino_12": "/content/azulino_12.jpeg", "azulino_13": "/content/azulino_13.jpeg", "beigs_09": "/content/beigs_09.jpeg", "beigs_10": "/content/beigs_10.jpeg", "beigs_07": "/content/beigs_07.jpeg", "beigs_06": "/content/beigs_06.jpeg", "beigs_08": "/content/beigs_08.jpeg", "beigs_05": "/content/beigs_05.jpeg", "marina_32": "/content/marina_32.jpeg", "marina_29": "/content/marina_29.jpeg", "marina_28": "/content/marina_28.jpeg", "marina_31": "/content/marina_31.jpeg", "marina_30": "/content/marina_30.jpeg", "rojo_26": "/content/rojo_26.jpeg", "rojo_23": "/content/rojo_23.jpeg", "rojo_21": "/content/rojo_21.jpeg", "rojo_22": "/content/rojo_22.jpeg", "rojo_25": "/content/rojo_25.jpeg", "rojo_24": "/content/rojo_24.jpeg", "celeste_40": "/content/celeste_40.jpeg", "celeste_38": "/content/celeste_38.jpeg", "celeste_39": "/content/celeste_39.jpeg", "celeste_33": "/content/celeste_33.jpeg", "celeste_35": "/content/celeste_35.jpeg", "celeste_37": "/content/celeste_37.jpeg", "celeste_41": "/content/celeste_41.jpeg", "celeste_42": "/content/celeste_42.jpeg", "celeste_34": "/content/celeste_34.jpeg", "celeste_36": "/content/celeste_36.jpeg", "sirenita_01": "/content/sirenita_01.png", "marino_18": "/content/marino_18.jpeg", "marino_17": "/content/marino_17.jpeg", "marino_16": "/content/marino_16.jpeg", "marino_15": "/content/marino_15.jpeg", "rosa_87": "/content/rosa_87.jpeg", "rosa_86": "/content/rosa_86.png", "rosa_79": "/content/rosa_79.jpeg", "rosa_82": "/content/rosa_82.png", "rosa_83": "/content/rosa_83.jpeg", "rosa_78": "/content/rosa_78.jpeg", "rosa_84": "/content/rosa_84.jpeg", "rosa_85": "/content/rosa_85.jpeg", "rosa_75": "/content/rosa_75.jpeg", "rosa_80": "/content/rosa_80.png", "rosa_81": "/content/rosa_81.png", "rosa_77": "/content/rosa_77.jpeg", "rosa_76": "/content/rosa_76.png", "blanco_55": "/content/blanco_55.jpeg", "blanco_56": "/content/blanco_56.jpeg", "blanco_53": "/content/blanco_53.jpeg", "blanco_52": "/content/blanco_52.jpeg", "blanco_57": "/content/blanco_57.jpeg", "blanco_49": "/content/blanco_49.jpeg", "blanco_51": "/content/blanco_51.jpeg", "blanco_60": "/content/blanco_60.jpeg", "blanco_47": "/content/blanco_47.jpeg", "blanco_44": "/content/blanco_44.jpeg", "blanco_50": "/content/blanco_50.jpeg", "blanco_48": "/content/blanco_48.jpeg", "blanco_59": "/content/blanco_59.jpeg", "blanco_43": "/content/blanco_43.jpeg", "blanco_58": "/content/blanco_58.png", "blanco_46": "/content/blanco_46.jpeg", "blanco_45": "/content/blanco_45.jpeg", "blanco_54": "/content/blanco_54.jpeg", } for tag in tags_detected: for key, path in tags_to_images.items(): if key in tag and check_image_exists(path): image_url = path break return full_response, image_url def update_image(image_url): if image_url: return image_url else: return None # Gradio layout setup with gr.Blocks() as demo: with gr.Row(): with gr.Column(scale=1): chatbot_input = gr.Textbox(label="Tu mensaje") chatbot_output = gr.Chatbot(label="ChatBot") chatbot_history = gr.State(value=[]) image_url = gr.State(value=None) submit_button = gr.Button("Enviar") with gr.Column(scale=1): image_output = gr.Image(label="Imagen asociada") def process_input(message, history): full_response, image = chat_function(message, history) history.append((message, full_response)) return history, history, image submit_button.click(process_input, inputs=[chatbot_input, chatbot_history], outputs=[chatbot_output, chatbot_history, image_url]) image_url.change(fn=update_image, inputs=image_url, outputs=image_output) # Launch the interface demo.launch(debug=True)