initial

README.md

@@ -1,13 +1,2 @@
----
-title: ResvuChatbox
-emoji: 💬
-colorFrom: yellow
-colorTo: purple
-sdk: gradio
-sdk_version: 4.36.1
-app_file: app.py
-pinned: false
-license: apache-2.0
----
-
-An example chatbot using [Gradio](https://gradio.app), [`huggingface_hub`](https://huggingface.co/docs/huggingface_hub/v0.22.2/en/index), and the [Hugging Face Inference API](https://huggingface.co/docs/api-inference/index).
+This is a Gradio-based chatbot interface for interacting with the Resvu API. Users can input messages and receive responses from the AI.
+

app.py

@@ -1,63 +1,42 @@
-import gradio as gr
-from huggingface_hub import InferenceClient
-
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
-
-
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-
-        response += token
-        yield response
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
-
-
-if __name__ == "__main__":
-    demo.launch()
+import subprocess
+import sys
+import gradio as gr
+from utils import initialize_database
+from model import llm_chain_response, get_response_value
+from process_documents import create_db_from_files
+
+def run_setup():
+    subprocess.check_call([sys.executable, "setup.py"])
+
+# Run the setup script
+run_setup()
+llm_chain = llm_chain_response()
+
+def chat_with_mistral(user_input):
+    if not user_input:
+      return "The message is not be empty."
+    response = llm_chain.invoke({"query": user_input})
+    print(response)
+
+    print("---------------Response--------------")
+    print(get_response_value(response["result"]))
+    return get_response_value(response["result"])
+
+def main():
+    # Initialize the database
+    create_db_from_files()
+
+    # Set up and launch the Gradio interface
+    iface = gr.Interface(
+        fn=chat_with_mistral,
+        inputs=gr.components.Textbox(label="Enter Your Message"),
+        outputs=gr.components.Markdown(label="ChatbotResponse"),
+        title="Resvu AI Chatbot",
+        description="Interact with the Resvu API via this chatbot. Enter a message and get a response.",
+        examples=["Hi, how are you", "Who are you?", "What services do you offer?", "How can I find out about upcoming community events?"],
+        allow_flagging="never"
+    )
+    iface.launch()
+
+if __name__ == "__main__":
+    main()

model.py

@@ -0,0 +1,118 @@
+from unsloth import FastLanguageModel
+import torch
+max_seq_length = 2048 # Choose any! We auto support RoPE Scaling internally!
+dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
+load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.
+
+from langchain_community.llms import CTransformers
+from langchain.chains import RetrievalQA
+from langchain.prompts import PromptTemplate
+from langchain_community.embeddings import GPT4AllEmbeddings
+from langchain_community.vectorstores import FAISS
+from langchain_community.llms import HuggingFacePipeline
+from langchain.callbacks.base import BaseCallbackHandler
+from transformers import pipeline
+
+# 4bit pre quantized models we support for 4x faster downloading + no OOMs.
+fourbit_models = [
+    "unsloth/mistral-7b-bnb-4bit",
+    "unsloth/mistral-7b-instruct-v0.2-bnb-4bit",
+    "unsloth/llama-2-7b-bnb-4bit",
+    "unsloth/llama-2-13b-bnb-4bit",
+    "unsloth/codellama-34b-bnb-4bit",
+    "unsloth/tinyllama-bnb-4bit",
+    "unsloth/gemma-7b-bnb-4bit", # New Google 6 trillion tokens model 2.5x faster!
+    "unsloth/gemma-2b-bnb-4bit",
+] # More models at https://huggingface.co/unsloth
+
+template = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
+
+### Instruction:
+You are ResVuAssist and You are a helpful bot who reads texts and answers questions about them.
+
+### Input:
+{context}
+QUESTION: {question}
+
+### Response:
+"""
+
+# Cau hinh
+vector_db_path = "vectorstores/db_faiss"
+
+
+def initialModelAndTokenizer():
+    model, tokenizer = FastLanguageModel.from_pretrained(
+        model_name = "unsloth/mistral-7b-instruct-v0.3-bnb-4bit", # Choose ANY! eg teknium/OpenHermes-2.5-Mistral-7B
+        max_seq_length = max_seq_length,
+        dtype = dtype,
+        load_in_4bit = load_in_4bit,
+        # token = "hf_...", # use one if using gated models like meta-llama/Llama-2-7b-hf
+    )
+    model = FastLanguageModel.get_peft_model(
+        model,
+        r = 16, # Choose any number > 0 ! Suggested 8, 16, 32, 64, 128
+        target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
+                        "gate_proj", "up_proj", "down_proj",],
+        lora_alpha = 16,
+        lora_dropout = 0, # Supports any, but = 0 is optimized
+        bias = "none",    # Supports any, but = "none" is optimized
+        # [NEW] "unsloth" uses 30% less VRAM, fits 2x larger batch sizes!
+        use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
+        random_state = 3407,
+        use_rslora = False,  # We support rank stabilized LoRA
+        loftq_config = None, # And LoftQ
+    )
+    return model, tokenizer
+
+def create_pipeline():
+    model, tokenizer = initialModelAndTokenizer()
+    pipe = pipeline(
+        "text-generation",
+        model=model,
+        tokenizer=tokenizer,
+        max_new_tokens=512,
+        temperature=0.1,
+        top_p=0.95,
+        repetition_penalty=1.15
+    )
+    return pipe
+
+# Tao prompt template
+def creat_prompt(template):
+    prompt = PromptTemplate(template = template, input_variables=["context", "question"])
+    return prompt
+
+# Tao simple chain
+def create_qa_chain(prompt, llm, db):
+    llm_chain = RetrievalQA.from_chain_type(
+        llm = llm,
+        chain_type= "stuff",
+        # retriever = db.as_retriever(search_kwargs = {"k":8}, max_tokens_limit=1024),
+        retriever = db.as_retriever(search_kwargs = {"k": 15}, max_tokens_limit=4096),
+        return_source_documents = False,
+        chain_type_kwargs= {'prompt': prompt},
+    )
+    return llm_chain
+
+# Read tu VectorDB
+def read_vectors_db():
+    # Embeding
+    embedding_model = GPT4AllEmbeddings(model_name="all-MiniLM-L6-v2.gguf2.f16.gguf")
+    db = FAISS.load_local(vector_db_path, embedding_model,  allow_dangerous_deserialization=True)
+    return db
+
+def get_response_value(text):
+    start = text.find('### Response:')
+    if start != -1:
+        return text[start + len('### Response:'):].strip()
+    return None
+
+def llm_chain_response():
+    pipe = create_pipeline()
+    db = read_vectors_db()
+    prompt = creat_prompt(template)
+    llm = HuggingFacePipeline(pipeline=pipe)
+
+    llm_chain  =create_qa_chain(prompt, llm, db)
+    return llm_chain

process-documents.py

@@ -0,0 +1,64 @@
+from langchain.text_splitter import RecursiveCharacterTextSplitter, CharacterTextSplitter
+from langchain_community.document_loaders import UnstructuredFileLoader, DirectoryLoader
+from langchain_community.vectorstores import FAISS
+from langchain_community.embeddings import GPT4AllEmbeddings
+
+# Khai bao bien
+pdf_data_path = "/data"
+vector_db_path = "vectorstores/db_faiss"
+
+# Ham 1. Tao ra vector DB tu 1 doan text
+def create_db_from_text():
+    raw_text = """
+      Thinh created you who is a chatbox at Resvu,
+      """
+
+    # Chia nho van ban
+    text_splitter = CharacterTextSplitter(
+        separator="\n",
+        chunk_size=100,
+        chunk_overlap=20,
+        length_function=len
+    )
+
+    chunks = text_splitter.split_text(raw_text)
+
+    # Embeding
+    embedding_model = GPT4AllEmbeddings(model_name="all-MiniLM-L6-v2.gguf2.f16.gguf")
+
+    # Dua vao Faiss Vector DB
+    db = FAISS.from_texts(texts=chunks, embedding=embedding_model)
+    db.save_local(vector_db_path)
+    return db
+
+# Define the file types you want to load
+file_types = ["*.pdf", "*.txt", "*.doc", "*.docx"]
+
+def create_db_from_files():
+    # Khai bao loader de quet toan bo thu muc dataa
+    # loader = DirectoryLoader(pdf_data_path, glob=file_types, loader_cls = PyPDFLoader)
+    # documents = loader.load()
+
+    # Create a loader for each file type
+    loaders = []
+    for file_type in file_types:
+        loader = DirectoryLoader(
+            pdf_data_path,
+            glob=file_type,
+            loader_cls=UnstructuredFileLoader
+        )
+        loaders.append(loader)
+
+    # Load all documents
+    documents = []
+    for loader in loaders:
+        documents.extend(loader.load())
+
+    text_splitter = RecursiveCharacterTextSplitter(chunk_size=200, chunk_overlap=50)
+    chunks = text_splitter.split_documents(documents)
+
+    # Embeding
+    embedding_model = GPT4AllEmbeddings(model_name="all-MiniLM-L6-v2.gguf2.f16.gguf")
+    db = FAISS.from_documents(chunks, embedding_model)
+    db.save_local(vector_db_path)
+    return db

requirements.txt

@@ -1 +1,12 @@
-huggingface_hub==0.22.2
+transformers
+langchain
+langchain-community
+torch
+pypdf
+sentence-transformers
+gpt4all
+faiss-cpu
+openai
+gradio
+unstructured
+python-docx

setup.py

@@ -0,0 +1,18 @@
+import subprocess
+import sys
+
+def install(package):
+    subprocess.check_call([sys.executable, "-m", "pip", "install", package])
+
+def setup_environment():
+    print("Installing required packages...")
+    install('git+https://github.com/unslothai/unsloth.git#egg=unsloth[colab-new]')
+    install('--no-deps xformers')
+    install('trl<0.9.0')
+    install('peft')
+    install('accelerate')
+    install('bitsandbytes')
+    print("Packages installed successfully.")
+
+if __name__ == "__main__":
+    setup_environment()