Alberto-Audrix committed
commit
32a36cceff
444 files changed
+67186
No files matched your search
@@ -0,0 +1,157 @@
|
||||
import os
|
||||
import sys
|
||||
import requests
|
||||
from dotenv import load_dotenv
|
||||
import chromadb
|
||||
from sentence_transformers import SentenceTransformer
|
||||
|
||||
# Set offline mode agar sentence-transformers tidak mencoba menghubungi Hugging Face di jaringan on-premise
|
||||
os.environ["HF_HUB_OFFLINE"] = "1"
|
||||
|
||||
# Load environment variables
|
||||
load_dotenv()
|
||||
|
||||
CHROMA_DB_DIR = os.getenv("CHROMA_DB_DIR", "chroma_db")
|
||||
EMBEDDING_MODEL_NAME = os.getenv("EMBEDDING_MODEL_NAME", "all-MiniLM-L6-v2")
|
||||
OPENAI_BASE_URL = os.getenv("OPENAI_BASE_URL", "http://localhost:1234/v1")
|
||||
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "lm-studio")
|
||||
LLM_MODEL_NAME = os.getenv("LLM_MODEL_NAME", "deepseek-r1-distill-qwen-7b")
|
||||
|
||||
def main():
|
||||
# 1. Inisialisasi Database Vektor
|
||||
if not os.path.exists(CHROMA_DB_DIR):
|
||||
print(f"Database Chroma DB di '{CHROMA_DB_DIR}' tidak ditemukan. Silakan jalankan extract_text.py dan populate_db.py terlebih dahulu.")
|
||||
return
|
||||
|
||||
print("Menghubungkan ke Chroma DB...")
|
||||
chroma_client = chromadb.PersistentClient(path=CHROMA_DB_DIR)
|
||||
try:
|
||||
collection = chroma_client.get_collection(name="company_sop")
|
||||
except Exception as e:
|
||||
print(f"Koleksi 'company_sop' tidak ditemukan di database. Pastikan populate_db.py sudah dijalankan dengan sukses. Error: {e}")
|
||||
return
|
||||
|
||||
# 2. Inisialisasi Model Embedding lokal
|
||||
print(f"Memuat model embedding lokal '{EMBEDDING_MODEL_NAME}' untuk kueri...")
|
||||
embedding_model = SentenceTransformer(EMBEDDING_MODEL_NAME)
|
||||
print("Model embedding berhasil dimuat.")
|
||||
|
||||
# 3. Konfigurasi koneksi LM Studio native v1 API
|
||||
LM_STUDIO_API_URL = os.getenv("LM_STUDIO_API_URL")
|
||||
if not LM_STUDIO_API_URL:
|
||||
openai_base = os.getenv("OPENAI_BASE_URL", "http://localhost:1234/v1")
|
||||
if openai_base.endswith("/v1"):
|
||||
LM_STUDIO_API_URL = openai_base.replace("/v1", "/api/v1/chat")
|
||||
else:
|
||||
LM_STUDIO_API_URL = f"{openai_base.rstrip('/')}/api/v1/chat"
|
||||
|
||||
print(f"Mengonfigurasi koneksi LLM ke native API: {LM_STUDIO_API_URL} (Model: {LLM_MODEL_NAME})...")
|
||||
|
||||
print("\n" + "="*60)
|
||||
print(" PIPELINE RAG LOKAL - ASISTEN SOP PERUSAHAAN (DEEPSEEK)")
|
||||
print(" Ketik 'keluar' atau 'exit' untuk menyudahi percakapan.")
|
||||
print("="*60 + "\n")
|
||||
|
||||
while True:
|
||||
try:
|
||||
query = input("\nPertanyaan Anda: ").strip()
|
||||
if not query:
|
||||
continue
|
||||
if query.lower() in ["keluar", "exit", "q", "quit"]:
|
||||
print("Sampai jumpa!")
|
||||
break
|
||||
|
||||
print("\n[1/3] Mencari dokumen referensi relevan di database lokal...", end="", flush=True)
|
||||
# Buat embedding kueri
|
||||
query_embedding = embedding_model.encode([query])[0].tolist()
|
||||
|
||||
# Cari kueri di Chroma DB (ambil 6 chunk teratas)
|
||||
results = collection.query(
|
||||
query_embeddings=[query_embedding],
|
||||
n_results=6
|
||||
)
|
||||
print(" Selesai!")
|
||||
|
||||
retrieved_chunks = results['documents'][0]
|
||||
retrieved_metadatas = results['metadatas'][0]
|
||||
|
||||
if not retrieved_chunks or len(retrieved_chunks) == 0:
|
||||
print("⚠️ Tidak ditemukan referensi dokumen yang cocok dengan pertanyaan Anda.")
|
||||
continue
|
||||
|
||||
# Tampilkan referensi yang ditemukan
|
||||
print("\n[Referensi yang Ditemukan]:")
|
||||
for idx, meta in enumerate(retrieved_metadatas):
|
||||
print(f" - [{idx+1}] File: {meta['source']} (Chunk: {meta['chunk_index']})")
|
||||
|
||||
# 4. Susun Prompt dengan Konteks SOP
|
||||
context = "\n\n---\n\n".join(retrieved_chunks)
|
||||
|
||||
system_prompt = (
|
||||
"Anda adalah asisten AI perusahaan yang profesional. Tugas Anda adalah memberikan jawaban "
|
||||
"yang valid, akurat, dan sesuai dengan Standar Operasional Prosedur (SOP) atau dokumen acuan perusahaan "
|
||||
"yang disediakan di bawah ini.\n"
|
||||
"Patuhi aturan berikut:\n"
|
||||
"1. Jawablah HANYA berdasarkan informasi yang ada dalam dokumen acuan di bawah.\n"
|
||||
"2. Jika jawaban tidak dapat ditemukan di dalam dokumen tersebut secara eksplisit atau logis, katakan dengan sopan "
|
||||
"bahwa 'Maaf, informasi tersebut tidak ditemukan dalam dokumen SOP/acuan perusahaan kami.' Jangan mengarang informasi.\n"
|
||||
"3. Sajikan data dengan valid dan rapi."
|
||||
)
|
||||
|
||||
user_prompt = f"""Dokumen SOP / Acuan Perusahaan:
|
||||
=========================================
|
||||
{context}
|
||||
=========================================
|
||||
|
||||
Pertanyaan Pengguna: {query}
|
||||
|
||||
Jawaban berdasarkan Dokumen Acuan:"""
|
||||
|
||||
print(f"\n[2/3] Menghubungi LLM DeepSeek lokal di {LM_STUDIO_API_URL}...", end="", flush=True)
|
||||
|
||||
# Panggil LM Studio native v1 API
|
||||
headers = {
|
||||
"Content-Type": "application/json"
|
||||
}
|
||||
if OPENAI_API_KEY and OPENAI_API_KEY != "lm-studio":
|
||||
headers["Authorization"] = f"Bearer {OPENAI_API_KEY}"
|
||||
|
||||
payload = {
|
||||
"model": LLM_MODEL_NAME,
|
||||
"input": user_prompt,
|
||||
"system_prompt": system_prompt,
|
||||
"temperature": 0.1,
|
||||
"max_output_tokens": 32000
|
||||
}
|
||||
|
||||
response = requests.post(LM_STUDIO_API_URL, json=payload, headers=headers)
|
||||
response.raise_for_status()
|
||||
print(" Selesai!")
|
||||
|
||||
answer = response.json()["response"]
|
||||
|
||||
# Pisahkan proses berpikir (<think>) jika ada (khusus model reasoning seperti DeepSeek R1)
|
||||
import re
|
||||
think_match = re.search(r'<think>(.*?)</think>', answer, re.DOTALL)
|
||||
clean_answer = re.sub(r'<think>.*?</think>', '', answer, flags=re.DOTALL).strip()
|
||||
|
||||
if think_match and think_match.group(1).strip():
|
||||
print(" Selesai!")
|
||||
print("\n[Proses Berpikir DeepSeek]:")
|
||||
print("." * 50)
|
||||
print(think_match.group(1).strip())
|
||||
print("." * 50)
|
||||
else:
|
||||
print(" Selesai!")
|
||||
|
||||
print("\n[3/3] Respon Asisten SOP:")
|
||||
print("-"*50)
|
||||
print(clean_answer)
|
||||
print("-"*50)
|
||||
|
||||
except Exception as e:
|
||||
print(f"\n❌ Terjadi kesalahan: {e}")
|
||||
print("Harap pastikan server LLM lokal Anda (LM Studio/vLLM/llama.cpp) sedang berjalan dan dapat diakses.")
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
Reference in new issue
Block a user