init project

.gitignore

@@ -0,0 +1,2 @@
+# files
+*.DS_Store

Dockerfile

@@ -0,0 +1,10 @@
+FROM python:3.10-slim
+
+WORKDIR /app
+COPY . .
+
+RUN pip install --upgrade pip \
+ && pip install -r requirements.txt
+
+EXPOSE 7860
+CMD ["python", "app.py"]

app.py

@@ -0,0 +1,45 @@
+import gradio as gr
+import logging
+from fastapi import FastAPI, Request
+from rag_core.chunker import chunk_legal_text
+from rag_core.embedder import get_embedding
+from rag_core.retriever import Retriever
+from rag_core.llm import generate_answer
+
+app = FastAPI()
+retriever = Retriever()
+
+# Khởi tạo nếu chưa có index
+if retriever.index is None:
+    logging.info("Không tìm thấy FAISS index, bắt đầu xử lý...")
+    with open("data/raw_law.txt", "r", encoding="utf-8") as f:
+        text = f.read()
+    chunks = chunk_legal_text(text)
+    retriever.build(chunks, get_embedding)
+
+# API endpoint
+@app.post("/ask")
+async def ask_api(req: Request):
+    data = await req.json()
+    query = data.get("query")
+    docs = retriever.query(query, get_embedding)
+    prompt = "\n\n".join(docs) + f"\n\nCâu hỏi: {query}\nTrả lời:"
+    answer = generate_answer(prompt)
+    return {"answer": answer}
+
+# Gradio UI
+iface = gr.Interface(
+    fn=lambda q: generate_answer("\n\n".join(retriever.query(q, get_embedding)) + f"\n\nCâu hỏi: {q}\nTrả lời:"),
+    inputs=gr.Textbox(label="Nhập câu hỏi"),
+    outputs=gr.Textbox(label="Trả lời"),
+    title="Luật Giao Thông RAG"
+)
+
+import uvicorn
+import threading
+
+def start_fastapi():
+    uvicorn.run(app, host="0.0.0.0", port=7861)
+
+threading.Thread(target=start_fastapi).start()
+iface.launch()

rag_core/chunker.py

@@ -0,0 +1,9 @@
+import re
+from typing import List
+from rag_core.utils import log_timed
+
+@log_timed("chunking văn bản luật")
+def chunk_legal_text(text: str) -> List[str]:
+    pattern = r"(Chương\\s+[IVXLC]+:.*?|Điều\\s+\\d+\\..*?)(?=(Chương\\s+[IVXLC]+:|Điều\\s+\\d+\\.|$))"
+    matches = re.findall(pattern, text, flags=re.DOTALL)
+    return [m[0].strip() for m in matches if len(m[0].strip()) > 30]

rag_core/embedder.py

@@ -0,0 +1,11 @@
+import requests
+from rag_core.utils import log_timed
+
+@log_timed("gửi API tạo embedding")
+def get_embedding(text: str):
+    response = requests.post(
+        "https://vietcat-phobertnode.hf.space/embed",
+        json={"text": text},
+        timeout=10
+    )
+    return response.json()["embedding"]

rag_core/llm.py

@@ -0,0 +1,11 @@
+import requests
+from rag_core.utils import log_timed
+
+@log_timed("gửi prompt tới LLM")
+def generate_answer(prompt: str) -> str:
+    response = requests.post(
+        "https://vietcat-gemma34b.hf.space/purechat",
+        json={"prompt": prompt},
+        timeout=30
+    )
+    return response.json()["response"]

rag_core/retriever.py

@@ -0,0 +1,35 @@
+import faiss
+import numpy as np
+import os
+import pickle
+from rag_core.utils import log_timed
+
+INDEX_PATH = "faiss_index/index.faiss"
+META_PATH = "faiss_index/meta.pkl"
+
+class Retriever:
+    def __init__(self):
+        if os.path.exists(INDEX_PATH):
+            self.index = faiss.read_index(INDEX_PATH)
+            with open(META_PATH, "rb") as f:
+                self.texts = pickle.load(f)
+        else:
+            self.index = None
+            self.texts = []
+
+    @log_timed("xây FAISS index")
+    def build(self, texts: list, embed_fn):
+        embeddings = [embed_fn(t) for t in texts]
+        dim = len(embeddings[0])
+        self.index = faiss.IndexFlatL2(dim)
+        self.index.add(np.array(embeddings).astype("float32"))
+        faiss.write_index(self.index, INDEX_PATH)
+        with open(META_PATH, "wb") as f:
+            pickle.dump(texts, f)
+        self.texts = texts
+
+    @log_timed("truy vấn FAISS")
+    def query(self, query_text, embed_fn, k=3):
+        q_emb = np.array([embed_fn(query_text)]).astype("float32")
+        D, I = self.index.search(q_emb, k)
+        return [self.texts[i] for i in I[0]]

rag_core/utils.py

@@ -0,0 +1,16 @@
+import time
+import logging
+
+logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(levelname)s] %(message)s")
+
+def log_timed(message):
+    def decorator(func):
+        def wrapper(*args, **kwargs):
+            logging.info(f"Bắt đầu {message}...")
+            start = time.time()
+            result = func(*args, **kwargs)
+            end = time.time()
+            logging.info(f"Hoàn tất {message} trong {end - start:.2f}s.")
+            return result
+        return wrapper
+    return decorator

requirements.txt

@@ -0,0 +1,6 @@
+faiss-cpu
+numpy
+requests
+gradio
+uvicorn
+fastapi