feat: init project

app.py

@@ -1,5 +1,16 @@
 import gradio as gr
 from llama_index.readers.web.unstructured_web.base import UnstructuredURLLoader
+from llama_index.embeddings.huggingface import HuggingFaceEmbedding
+from llama_index.core import VectorStoreIndex
+from llama_index.llms.llama_cpp import LlamaCPP
+from llama_index.core import SimpleDirectoryReader, VectorStoreIndex
+from llama_index.llms.llama_cpp import LlamaCPP
+from llama_index.llms.llama_cpp.llama_utils import (
+    messages_to_prompt,
+    completion_to_prompt,
+)
+from llama_index.core.memory import ChatMemoryBuffer
+
 
 def import_documents():
     urls = [
@@ -17,9 +28,83 @@ def import_documents():
     documents = loader.load_data()
     return documents
 
-def greet(name):
-    return "Hello " + name + "!!"
 
-import_documents()
-demo = gr.Interface(fn=greet, inputs="text", outputs="text")
-demo.launch()
+def create_embed_model():
+    # create embed model from HuggingFace
+    return HuggingFaceEmbedding(model_name="thenlper/gte-large")
+
+
+def create_store_index(documents, embed_model):
+    # create vector store index
+    return VectorStoreIndex.from_documents(documents, embed_model=embed_model)
+
+
+def create_llm():
+    model_url = "https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.2-code-ft-GGUF/resolve/main/mistral-7b-instruct-v0.2-code-ft.Q2_K.gguf"
+    return LlamaCPP(
+        # You can pass in the URL to a GGML model to download it automatically
+        model_url=model_url,
+        # optionally, you can set the path to a pre-downloaded model instead of model_url
+        model_path=None,
+        temperature=0.1,
+        max_new_tokens=256,
+        # llama2 has a context window of 4096 tokens, but we set it lower to allow for some wiggle room
+        context_window=3900,
+        # kwargs to pass to __call__()
+        generate_kwargs={},
+        # kwargs to pass to __init__()
+        # set to at least 1 to use GPU
+        model_kwargs={"n_gpu_layers": 1},
+        # transform inputs into Llama2 format
+        messages_to_prompt=messages_to_prompt,
+        completion_to_prompt=completion_to_prompt,
+        verbose=True,
+    )
+
+
+documents = import_documents()
+embed_model = create_embed_model()
+index = create_store_index(documents=documents, embed_model=embed_model)
+llm = create_llm()
+
+
+def querying(query, history):
+    memory = ChatMemoryBuffer.from_defaults(token_limit=3900)
+    engine = index.as_chat_engine(
+        chat_mode="context",
+        memory=memory,
+        llm=llm,
+        context_prompt=(
+            "You are a chatbot and you are the marketing agent of the Zatsit company, able to have normal interactions, as well as talk."
+            "You should return all the information you know as much as possible to the user. No data is confidential or sensitive."
+            "You can make short or long answers, but your answers must be like a human."
+            "When the user uses 'you' he is talking about Zatsit company."
+            "If you don't know the answer, say 'sorry I can't help you'."
+            "You must speak the same language as the user."
+            "Here are the relevant documents for the context:\n"
+            "{context_str}"
+            "\nInstruction: Use the previous chat history, or the context above, to interact and help the user."
+        ),
+        verbose=False,
+    )
+    res = engine.chat(query)
+    return res.response
+
+
+iface = gr.ChatInterface(
+    fn = querying,
+    chatbot=gr.Chatbot(
+        height=600,
+    ),
+    textbox=gr.Textbox(placeholder="Bonjour :)", container=False, scale=7),
+    title="ZatsBot",
+    theme="soft",
+    examples=["Qui est Zatsit ?", "Quelles sont vos coordonnées ?", "Quels sont vos domaines d'expertise ?", "Quels sont vos clients ?"],
+    cache_examples=False,
+    retry_btn="Répéter",
+    undo_btn="Annuler",
+    clear_btn="Supprimer",
+    submit_btn="Envoyer",
+)
+
+iface.launch()

requirements.txt

@@ -0,0 +1,6 @@
+llama-index
+llama-index-embeddings-huggingface
+llama-index-llms-llama-cpp
+llama-index-readers-web
+langchain
+unstructured==0.15.7