import nest_asyncio nest_asyncio.apply() from llama_index.core import ( VectorStoreIndex, ServiceContext, SimpleDirectoryReader, load_index_from_storage, ) from llama_index.core.storage import StorageContext from llama_index.core.node_parser import SentenceSplitter from llama_index.core.prompts import PromptTemplate from llama_index.core.response_synthesizers import TreeSummarize from llama_index.core.query_pipeline import InputComponent from llama_index.core.indices.knowledge_graph import KGTableRetriever from llama_index.legacy.vector_stores.faiss import FaissVectorStore from llama_index.llms.openai import OpenAI from llama_index.embeddings.openai import OpenAIEmbedding from llama_index.core import Settings import openai import os from github import Github from datetime import datetime import gradio as gr # OpenAI: openai.api_key = os.environ.get('openai_key') os.environ["OPENAI_API_KEY"] = os.environ.get('openai_key') # Github: exec(os.environ.get('logs_context')) # Context: exec(os.environ.get('context')) project_name = "DEV PharmaWise Data Integrity Chat 4.5" import networkx as nx import matplotlib.pyplot as plt from PIL import Image from io import BytesIO def draw_graph(): global kg_data G = nx.DiGraph() for source, relation, target in kg_data: G.add_edge(source, target, label=relation) # Utilizar spring_layout para mejorar la disposición de los nodos pos = nx.spring_layout(G) plt.figure(figsize=(12, 8)) # Ajustar el tamaño de los nodos nx.draw(G, pos, with_labels=True, node_color='skyblue', node_size=400, edge_color='k', linewidths=1, font_size=8, font_weight='bold') # Ajustar el tamaño de las flechas y el espaciado entre ellas edge_labels = {} for source, target, data in G.edges(data=True): if 'label' in data: edge_labels[(source, target)] = data['label'] nx.draw_networkx_edge_labels(G, pos, edge_labels=edge_labels, font_size=7, font_weight='normal') plt.title("Graph") plt.axis('off') buf = BytesIO() plt.savefig(buf, format='png') buf.seek(0) plt.close() return Image.open(buf) def extraer_informacion_metadata(respuesta, max_results=10): # Obtener source_nodes de la respuesta source_nodes = respuesta.source_nodes # Obtener page_labels, file_names y scores de source_nodes page_file_info = [ f"Página {node.node.metadata.get('page_label', '')} del archivo {node.node.metadata.get('file_name', '')} (Relevance: {node.score:.6f} - Id: {node.node.id_})" for node in source_nodes ] # Limitar la cantidad de resultados page_file_info = page_file_info[:max_results] return page_file_info from typing import List from llama_index.core import Prompt from llama_index.core.llms import ChatMessage, MessageRole from llama_index.core.chat_engine.context import ContextChatEngine from llama_index.core.memory import ChatMemoryBuffer chat_history_engine = [] result_metadata = "" with gr.Blocks(theme='sudeepshouche/minimalist') as demo: def visible(): return {btn_graph: gr.Button(value="Grafo", visible=True)} def get_ref(): return {mkdn: gr.Markdown(result_metadata)} def refresh(chat_history): global kg_data global chat_history_engine global result_metadata kg_data = [] chat_history_engine = [] result_metadata = "" chat_history = [[None, None]] gr.Info("¡Listo! Ya puedes seguir chateando.") return chat_history def summarize_assistant_messages(chat_history: List[ChatMessage]) -> List[ChatMessage]: # Encontrar la anteúltima respuesta del asistente assistant_messages = [msg for msg in chat_history if msg.role == MessageRole.ASSISTANT] if len(assistant_messages) < 2: return chat_history # No hay suficientes mensajes del asistente para resumir anteultima_respuesta = assistant_messages[-2] # Usar GPT-3.5 para generar un resumen de la anteúltima respuesta del asistente prompt = Prompt(f"Responder SOLO con un resumen del siguiente texto: \n\n{anteultima_respuesta.content}") llm = OpenAI(model="gpt-3.5-turbo", temperature=0.1) response = llm.predict(prompt) # Crear un nuevo ChatMessage con el resumen como contenido y el rol de asistente summarized_message = ChatMessage(content=response, role=MessageRole.ASSISTANT) # Reconstruir el historial de chat reemplazando la anteúltima respuesta del asistente con el resumen new_chat_history = [msg if msg != anteultima_respuesta else summarized_message for msg in chat_history] return new_chat_history def respond(message, chat_history): global chat_history_engine global result_metadata # Si chat_history está vacío, inicialízalo con el mensaje del usuario actual if not chat_history: chat_history = [[message, ""]] else: # Si chat_history no está vacío, agrega el mensaje actual al final de la lista chat_history.append([message, ""]) chat_history_engine = summarize_assistant_messages(chat_history_engine) #chat_history_engine.append(ChatMessage(role=MessageRole.USER, content=message)) response = chat_engine.stream_chat(message, chat_history=chat_history_engine) # Extraer información de source_nodes metadata_info = extraer_informacion_metadata(response, max_results=10) # Presentar la información de source_nodes en forma de lista con bullets if metadata_info: metadata_list = "\n".join(["- " + info for info in metadata_info]) result_metadata = "\n\n" + metadata_list for text in response.response_gen: chat_history[-1][1] += text # Agrega el texto de respuesta al último mensaje en chat_history yield "", chat_history print("----------") print(memory.get_all()) #chat_history_engine.append(ChatMessage(role=MessageRole.ASSISTANT, content=chat_history[-1][1])) #return "", chat_history gr.Markdown(""" # PharmaWise Data Integrity Chat 4.5 Realiza preguntas a tus datos y obtén al final del texto las paginas y documentos utilizados generar tu responder. """) with gr.Row(): with gr.Column(): chatbot = gr.Chatbot(show_label=False, show_copy_button=True, ) #layout="panel" pregunta = gr.Textbox(show_label=False, autofocus=True, placeholder="Realiza tu consulta...") pregunta.submit(respond, [pregunta, chatbot], [pregunta, chatbot]) with gr.Row(): btn_send = gr.Button(value="Preguntar", variant="primary") clear = gr.Button(value="Limpiar") gr.Examples(label="Ejemplos", examples=["Explicar el concepto ALCOA"], inputs=[pregunta]) with gr.Column(): with gr.Accordion(label="Bases de datos del conocimiento", open=False): gr.Markdown(""" ###### [1] ISPE Risk Based Approach to Compliant Electronic Records and Signatures.pdf ###### [2] EMA Guideline on computerised systems and electronic data in clinical trials.pdf ###### [3] EU GMP guide annexes Supplementary requirements Annex 11 Computerised systems.pdf ###### [4] FDA Data Integrity and Compliance With Drug CGMP (Q&A) Guidance for Industry.pdf ###### [5] GAMP 5 A Risk Based Approach to Compliant GxP Computerized System (ED2).pdf ###### [6] ISPE Application of GAMP 5 to Implementation of a GxP Clinical System.pdf ###### [7] ISPE Guide_ Project Management for the Pharmaceutical Industry - ISPE.pdf ###### [8] ISPE Science and Risk-Based Approach for the Delivery of Facilities, Systems, and Equipment.pdf ###### [9] GAMP Good Practice Guide The Validation of Legacy Systems.pdf ###### [10] ISPE Manufacturing Execution Systems.pdf ###### [11] MHRA GXP Data Integrity Guidance and Definitions (2018).pdf ###### [12] PI 041-1 Good Practices for Data Management and Integrity in Regulated Environments (2021).pdf ###### [13] ISPE Records and Data Integrity Guide.pdf ###### [14] ISPE Testing GxP Systems.pdf ###### [15] WHO TR 1033 Annex 4 Guideline on data integrity.pdf ###### [16] ISPE Validation of Laboratory Computerized Systems 2005.pdf ###### [17] FDA Guidance for Industry Part 11, Electronic Records; Electronic Signatures.pdf ###### [18] FDA General Principles of Software Validation -2002.pdf ###### [19] FDA Guidance for Industry Computerized Systems Used in Clinical Trials.pdf ###### [20] FDA Guidance-Computer-Software-Assurance - 2022.pdf ###### [21] FDA 21 CFR Part 11 Electronic Records Electronic-Signatures - 2003.pdf ###### [22] EMA Annex 15 Qualification and Validation.pdf ###### [23] EMA Annex 11 Computerised Systems.pdf ###### [24] ANMAT Disposicion_3827-2018 - Anexo 6 Sistemas Informaticos.pdf ###### [25] DIGEMID DS-021 5.6 Sistemas Computadorizados.pdf ###### [26] COFEPRIS BPM-NOM059-2015 9.13 Validación de sistemas computacionales.pdf ###### [27] PA-PH-OMCL (08) 88 R5 Annex 2 Validation of Complex Computerised Systems.pdf ###### [28] INVIMA Resolución_3619 - GLP- 2013 5 Equipos procesadores de datos.pdf ###### [29] PAPHOMCL (08) 87 R6 Annex 1 Validation of Excel Spreadsheets.pdf ###### [30] PA-PH-OMCL (08) 69 R7 Validation of Computerised Systems.pdf ###### [31] GUÍA BPM ARCSA 2020 (software).pdf ###### [32] GMP Paraguay DINAVISA Resolucion 197-21.pdf ###### [33] ANVISA Guide for Computer Systems Validation 33-2020.pdf ###### [34] ANVISA INSTRUÇÃO NORMATIVA - IN Nº 43 - 2019.pdf ###### [35] PIC_011_3_recommendation_on_computerised_systems.pdf ###### [36] PIC Revision of Annex 11 EU GMP.pdf """) with gr.Accordion(label="Referencias", open=True): mkdn = gr.Markdown() with gr.Row(): btn_graph = gr.Button(value="Grafo") btn_ref = gr.Button(value="Referencias") with gr.Row(): grafo = gr.Image(label="Grafo", show_share_button=False) btn_ref.click(get_ref, outputs=[mkdn]) btn_send.click(respond, [pregunta, chatbot], [pregunta, chatbot]) btn_graph.click(draw_graph, outputs=[grafo]) clear.click(refresh, inputs=[chatbot], outputs=[chatbot]) #response.change(visible, [], [btn_graph]) demo.queue() demo.launch()