Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| from huggingface_hub import InferenceClient | |
| from transformers import AutoTokenizer, AutoModel | |
| from langchain.text_splitter import RecursiveCharacterTextSplitter | |
| from langchain_huggingface import HuggingFaceEmbeddings | |
| from langchain_community.vectorstores import FAISS | |
| import fitz # PyMuPDF | |
| def get_hf_models(): | |
| return ["Qwen/Qwen2.5-3B-Instruct", "HuggingFaceH4/zephyr-7b-beta", "mistralai/Mistral-7B-Instruct-v0.1"] | |
| def extract_text_from_pdf(pdf_path): | |
| text = "" | |
| with fitz.open(pdf_path) as doc: | |
| for page in doc: | |
| text += page.get_text() | |
| return text | |
| def manual_rag(query, context, client): | |
| prompt = f"Context: {context}\n\nQuestion: {query}\n\nAnswer:" | |
| response = client.text_generation(prompt, max_new_tokens=512) | |
| return response | |
| def classic_rag(query, pdf_path, client, embedder): | |
| text = extract_text_from_pdf(pdf_path) | |
| text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200) | |
| chunks = text_splitter.split_text(text) | |
| embeddings = HuggingFaceEmbeddings(model_name=embedder) | |
| db = FAISS.from_texts(chunks, embeddings) | |
| docs = db.similarity_search(query, k=3) | |
| context = " ".join([doc.page_content for doc in docs]) | |
| response = manual_rag(query, context, client) | |
| return response, context | |
| def no_rag(query, client): | |
| response = client.text_generation(query, max_new_tokens=512) | |
| return response | |
| def process_query(query, pdf_path, llm_choice, embedder_choice, use_manual_rag): | |
| client = InferenceClient(llm_choice) | |
| no_rag_response = no_rag(query, client) | |
| if pdf_path is None: | |
| return no_rag_response, "RAG non utilisé (pas de fichier PDF)", "RAG non utilisé (pas de fichier PDF)", "Pas de fichier PDF fourni", "Pas de contexte extrait" | |
| full_text = extract_text_from_pdf(pdf_path) | |
| # RAG manuel seulement si choisi | |
| if use_manual_rag == "Oui": | |
| manual_rag_response = manual_rag(query, full_text, client) | |
| else: | |
| manual_rag_response = "RAG manuel non utilisé" | |
| classic_rag_response, classic_rag_context = classic_rag(query, pdf_path, client, embedder_choice) | |
| return no_rag_response, manual_rag_response, classic_rag_response, full_text, classic_rag_context | |
| iface = gr.Interface( | |
| fn=process_query, | |
| inputs=[ | |
| gr.Textbox(label="Votre question"), | |
| gr.File(label="Chargez un nouveau PDF"), | |
| gr.Dropdown(choices=get_hf_models(), label="Choisissez le LLM", value="HuggingFaceH4/zephyr-7b-beta"), | |
| gr.Dropdown(choices=["sentence-transformers/all-MiniLM-L6-v2", "nomic-ai/nomic-embed-text-v1.5"], | |
| label="Choisissez l'Embedder", value="sentence-transformers/all-MiniLM-L6-v2"), | |
| gr.Dropdown(choices=["Oui", "Non"], label="Utiliser RAG manuel ?", value="Non") # Ajout de la combobox pour choisir RAG manuel | |
| ], | |
| outputs=[ | |
| gr.Textbox(label="Réponse sans RAG"), | |
| gr.Textbox(label="Réponse avec RAG manuel"), | |
| gr.Textbox(label="Réponse avec RAG classique"), | |
| gr.Textbox(label="Texte complet du PDF (pour RAG manuel)", lines=10), | |
| gr.Textbox(label="Contexte extrait (pour RAG classique)", lines=10) | |
| ], | |
| title="Tutoriel RAG - Comparaison des méthodes", | |
| description="Posez une question sur le contenu d'un PDF et comparez les réponses obtenues avec différentes méthodes de RAG.", | |
| theme="default" | |
| ) | |
| if __name__ == "__main__": | |
| iface.launch() | |