import gradio as gr from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer import torch import theme theme = theme.Theme() import pydantic import os import sys sys.path.append('../..') DEVEL = os.environ.get('DEVEL', False) #langchain from langchain.text_splitter import RecursiveCharacterTextSplitter, CharacterTextSplitter from langchain.embeddings import HuggingFaceEmbeddings from langchain.prompts import PromptTemplate from langchain.chains import RetrievalQA from langchain.prompts import ChatPromptTemplate from langchain.schema import StrOutputParser from langchain.schema.runnable import Runnable from langchain.schema.runnable.config import RunnableConfig from langchain.chains import ( LLMChain, ConversationalRetrievalChain) from langchain.vectorstores import Chroma from langchain.memory import ConversationBufferMemory from langchain.chains import LLMChain from langchain.prompts.prompt import PromptTemplate from langchain.prompts.chat import ChatPromptTemplate, SystemMessagePromptTemplate from langchain.prompts import SystemMessagePromptTemplate, HumanMessagePromptTemplate, ChatPromptTemplate, MessagesPlaceholder from langchain.document_loaders import PyPDFDirectoryLoader from langchain.pydantic_v1 import BaseModel, Field, validator from langchain.output_parsers import PydanticOutputParser from langchain_community.llms import HuggingFaceHub from langchain_community.document_loaders import WebBaseLoader from typing import List from langchain.llms import OpenAI from langchain.output_parsers import PydanticOutputParser from langchain.prompts import PromptTemplate from pydantic import BaseModel, Field import shutil title= "Green Greta" from huggingface_hub import from_pretrained_keras import tensorflow as tf from tensorflow import keras from PIL import Image # Cell 1: Image Classification Model model1 = from_pretrained_keras("rocioadlc/EfficientNetV2L") # Define class labels class_labels = ['battery', 'biological', 'brown-glass', 'cardboard', 'clothes', 'green-glass', 'metal', 'paper', 'plastic', 'shoes', 'trash', 'white-glass'] def predict_image(image_input): # Resize the image to the size expected by the model image = image_input.resize((244, 224)) # Convert the image to a NumPy array image_array = tf.keras.preprocessing.image.img_to_array(image) # Normalize the image image_array /= 255.0 # Expand the dimensions to create a batch image_array = tf.expand_dims(image_array, 0) # Predict using the model predictions = model1.predict(image_array) category_scores = {} for i, class_label in enumerate(class_labels): category_scores[class_label] = predictions[0][i].item() return category_scores #Cell 2: Chatbot loader = WebBaseLoader(["https://www.epa.gov/recycle/frequent-questions-recycling", "https://www.whitehorsedc.gov.uk/vale-of-white-horse-district-council/recycling-rubbish-and-waste/lets-get-real-about-recycling/", "https://www.teimas.com/blog/13-preguntas-y-respuestas-sobre-la-ley-de-residuos-07-2022", "https://www.molok.com/es/blog/gestion-de-residuos-solidos-urbanos-rsu-10-dudas-comunes"]) data=loader.load() # split documents text_splitter = RecursiveCharacterTextSplitter( chunk_size=1024, chunk_overlap=150, length_function=len ) docs = text_splitter.split_documents(data) # define embedding embeddings = HuggingFaceEmbeddings(model_name='thenlper/gte-small') # create vector database from data persist_directory = 'docs/chroma/' # Remove old database files if any shutil.rmtree(persist_directory, ignore_errors=True) vectordb = Chroma.from_documents( documents=docs, embedding=embeddings, persist_directory=persist_directory ) # define retriever retriever = vectordb.as_retriever(search_kwargs={"k": 2}, search_type="mmr") class Answer(BaseModel): question: str = Field(description="the original question") answer: str = Field(description="the extracted answer") # Set up a PydanticOutputParser parser = PydanticOutputParser(pydantic_object=Answer) # Create a prompt with format instructions prompt = PromptTemplate( template="Answer the user query.\n{format_instructions}\n{query}\n", input_variables=["query"], partial_variables={"format_instructions": parser.get_format_instructions()}, ) template = """ Your name is Greta and you are a recycling chatbot with the objective to anwer questions from user in English or Spanish / Use the following pieces of context to answer the question / If the question is English answer in English / If the question is Spanish answer in Spanish / Do not mention the word context when you answer a question / Answer the question fully and provide as much relevant detail as possible. Do not cut your response short / Context: {context} User: {question} {format_instructions} """ # Create the chat prompt templates sys_prompt = SystemMessagePromptTemplate.from_template(template) qa_prompt = ChatPromptTemplate( messages=[ sys_prompt, HumanMessagePromptTemplate.from_template("{question}")], partial_variables={"format_instructions": parser.get_format_instructions()} ) llm = HuggingFaceHub( repo_id="mistralai/Mixtral-8x7B-Instruct-v0.1", task="text-generation", model_kwargs={ "max_new_tokens": 2000, "top_k": 30, "temperature": 0.1, "repetition_penalty": 1.03 }, ) qa_chain = ConversationalRetrievalChain.from_llm( llm = llm, memory = ConversationBufferMemory(llm=llm, memory_key="chat_history", input_key='question', output_key='output'), retriever = retriever, verbose = True, combine_docs_chain_kwargs={'prompt': qa_prompt}, get_chat_history = lambda h : h, rephrase_question = False, output_key = 'output', ) def chat_interface(question,history): result = qa_chain.invoke({'question': question}) output_string = result['output'] # Find the index of the last occurrence of "answer": in the string answer_index = output_string.rfind('"answer":') # Extract the substring starting from the "answer": index answer_part = output_string[answer_index + len('"answer":'):].strip() # Find the next occurrence of a double quote to get the start of the answer value quote_index = answer_part.find('"') # Extract the answer value between double quotes answer_value = answer_part[quote_index + 1:answer_part.find('"', quote_index + 1)] return answer_value image_gradio_app = gr.Interface( fn=predict_image, inputs=gr.Image(label="Image", sources=['upload', 'webcam'], type="pil"), outputs=[gr.Label(label="Result")], title="Green Greta", theme=theme ) chatbot_gradio_app = gr.ChatInterface( fn=chat_interface, title="Green Greta" ) banner_tab_content = """
Banner Image

¡Bienvenido a nuestro clasificador de imágenes y chatbot para un reciclaje más inteligente!♻️

¿Alguna vez te has preguntado si puedes reciclar un objeto en particular? ¿O te has sentido abrumado por la cantidad de residuos que generas y no sabes cómo manejarlos de manera más sostenible? ¡Estás en el lugar correcto!

Nuestra plataforma combina la potencia de la inteligencia artificial con la comodidad de un chatbot para brindarte respuestas rápidas y precisas sobre qué objetos son reciclables y cómo hacerlo de la manera más eficiente.

¿Cómo usarlo?

""" banner_tab = gr.Markdown(banner_tab_content) # Combinar ambas interfaces en una sola aplicación con pestañas app = gr.TabbedInterface( [banner_tab, image_gradio_app, chatbot_gradio_app], tab_names=["Welcome to Green Greta", "Green Greta Image Classification", "Green Greta Chat"], theme=theme ) app.queue() app.launch()