File size: 4,680 Bytes
23bdbfd |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 |
import os
from huggingface_hub import login
from datasets import load_dataset
import gradio as gr
from llama_cpp import Llama
from huggingface_hub import hf_hub_download
import chromadb
from sentence_transformers import SentenceTransformer
from dotenv import load_dotenv
# Charger le fichier .env
load_dotenv()
# Lire le token depuis l'environnement
hf_token = os.getenv("HF_TOKEN")
# Authentification via un secret
hf_token = os.getenv("HF_TOKEN") # Récupérer le token depuis les secrets
login(hf_token)
# Charger le dataset
dataset = load_dataset("Maryem2025/dataset-train") # Changez le nom si nécessaire
# Initialisation du modèle Llama
llm = Llama(
model_path=hf_hub_download(
repo_id="TheBloke/CapybaraHermes-2.5-Mistral-7B-GGUF",
filename="capybarahermes-2.5-mistral-7b.Q2_K.gguf",
),
n_ctx=2048,
n_gpu_layers=50, # Ajustez selon votre VRAM
)
# Initialisation de ChromaDB Vector Store
class VectorStore:
def __init__(self, collection_name):
self.embedding_model = SentenceTransformer('sentence-transformers/multi-qa-MiniLM-L6-cos-v1')
self.chroma_client = chromadb.Client()
# Supprimer la collection existante si elle existe
if collection_name in self.chroma_client.list_collections():
self.chroma_client.delete_collection(collection_name)
# Créer une nouvelle collection
self.collection = self.chroma_client.create_collection(name=collection_name)
def populate_vectors(self, dataset):
# Sélectionner les colonnes pertinentes à concaténer
names = dataset['train']['name'][:20]
ingredients = dataset['train']['ingredients'][:20]
instructions = dataset['train']['instructions'][:20]
cuisine = dataset['train']['cuisine'][:20]
total_time = dataset['train']['total_time'][:20]
# Concaténer les textes à partir des colonnes sélectionnées
texts = [
f"Name: {name}. Ingredients: {ingr}. Instructions: {instr}. Cuisine: {cui}. Total time: {total} minutes."
for name, ingr, instr, cui, total in zip(names, ingredients, instructions, cuisine, total_time)
]
# Ajouter les embeddings au store de vecteurs
for i, item in enumerate(texts):
embeddings = self.embedding_model.encode(item).tolist()
self.collection.add(embeddings=[embeddings], documents=[item], ids=[str(i)])
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']
# Initialisation du store de vecteurs et peuplement
dataset = load_dataset('Maryem2025/dataset-test')
vector_store = VectorStore("embedding_vector")
vector_store.populate_vectors(dataset)
# Fonction pour générer du texte
def generate_text(message, max_tokens, temperature, top_p):
# Récupérer le contexte depuis le store de vecteurs
context_results = vector_store.search_context(message, n_results=1)
context = context_results[0] if context_results else ""
# Créer le modèle de prompt
prompt_template = (
f"SYSTEM: You are a recipe generating bot.\n"
f"SYSTEM: {context}\n"
f"USER: {message}\n"
f"ASSISTANT:\n"
)
# Générer le texte avec le modèle de langue
output = llm(
prompt_template,
temperature=0.3,
top_p=0.95,
top_k=40,
repeat_penalty=1.1,
max_tokens=600,
)
# Traiter la sortie
input_string = output['choices'][0]['text'].strip()
cleaned_text = input_string.strip("[]'").replace('\\n', '\n')
continuous_text = '\n'.join(cleaned_text.split('\n'))
return continuous_text
# Définir l'interface Gradio
demo = gr.Interface(
fn=generate_text,
inputs=[
gr.Textbox(lines=2, placeholder="Enter your message here...", label="Message"),
],
outputs=gr.Textbox(label="Generated Text"),
title="Chatbot - Your Personal Culinary Advisor: Discover What to Cook Next!",
description="Running LLM with context retrieval from ChromaDB",
examples=[
["I have leftover rice, what can I make out of it?"],
["I just have some milk and chocolate, what dessert can I make?"],
["I am allergic to coconut milk, what can I use instead in a Thai curry?"],
["Can you suggest a vegan breakfast recipe?"],
["How do I make a perfect scrambled egg?"],
["Can you guide me through making a soufflé?"],
],
cache_examples=False,
)
if __name__ == "__main__":
demo.launch()
|