new zephyr model

app.py

@@ -1,5 +1,38 @@
 import gradio as gr
 import requests
+import os
+from pathlib import Path
+from llama_index.embeddings import HuggingFaceEmbedding, VoyageEmbedding
+from llama_index import (load_index_from_storage, ServiceContext, StorageContext, VectorStoreIndex)
+from llama_index import download_loader, SimpleDirectoryReader
+from llama_index.retrievers import RecursiveRetriever
+from llama_index.query_engine import RetrieverQueryEngine
+from llama_index.llms import Anyscale
+
+# Define the inference model
+# llm = Anyscale(model="mistralai/Mistral-7B-Instruct-v0.1", api_key=os.getenv("ANYSCALE_API_KEY"))
+llm = Anyscale(model="HuggingFaceH4/zephyr-7b-beta", api_key=os.getenv("ANYSCALE_API_KEY"))
+# Define the embedding model used to embed the query.
+# query_embed_model = HuggingFaceEmbedding(model_name="BAAI/bge-base-en-v1.5")
+embed_model = VoyageEmbedding(model_name="voyage-01", voyage_api_key=os.getenv("VOYAGE_API_KEY"))
+service_context = ServiceContext.from_defaults(llm=llm, embed_model=embed_model)
+
+if "index" in os.listdir():
+    storage_context = StorageContext.from_defaults(persist_dir=Path("./index"))
+else: 
+    dir_reader = SimpleDirectoryReader(Path('./docs'))
+    documents = dir_reader.load_data()
+    index = VectorStoreIndex.from_documents(documents, service_context=service_context)
+    index.storage_context.persist(Path('./index'))
+    storage_context = StorageContext.from_defaults(persist_dir=Path("./index"))
+
+# Load the vector stores that were created earlier.
+index = load_index_from_storage(storage_context=storage_context, service_context=service_context)   
+
+# Define query engine:
+index_engine = index.as_retriever(similarity_top_k=4)
+index_retriever = RecursiveRetriever("vector",retriever_dict={"vector": index_engine})
+query_engine = RetrieverQueryEngine.from_args(index_retriever, service_context=service_context) 
 
 def authenticate(username, password):
     if username == "Gribouille" and password == "A jamais les premiers":
@@ -8,8 +41,7 @@ def authenticate(username, password):
         return False
 
 def predict(query):
-    response = requests.get(f"https://pbrenotvinciguerra-corte-api.hf.space/generate?query={query}") 
-    return str(response.text)
+    return str(query_engine.query(query))
 
 iface = gr.Interface(fn=predict, inputs=["text"], outputs="text")
 iface.launch(auth=authenticate)