llmproccrs / config.py
aitek230telu's picture
Upload 3 files
eab2784 verified
Raw History Blame Contribute Delete
2 kB
import os
import torch
from langchain_openai import ChatOpenAI
class Config:
# --- LLM Settings ---
# API_KEY = "sk-4GMsTHzzKO97UIIqq9RfEQ"
# BASE_URL = "https://dekallm.cloudeka.ai/v1"
# MODEL_NAME = "qwen/qwen25-72b-instruct"
# --- LLM CONFIGURATION (CLOUDEKA) ---
# Masukkan API Key Anda di sini
API_KEY = "sk-4GMsTHzzKO97UIIqq9RfEQ"
# Base URL CloudEka
BASE_URL = "http://100.90.107.65:1234/v1"
# Model Name
# MODEL_NAME = "mistralai/ministral-3-3b"
MODEL_NAME = "qwen/qwen2.5-vl-7b"
# MODEL_NAME = "google/gemma-3-4b"
# MODEL_NAME = "openai/gpt-oss-20b"
# --- System Settings ---
MAX_CANDIDATES_FOR_RECOM = 10 # Batas untuk switch dari 'Asking' ke 'Recommendation'
# === PATHS ===
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
DATASET_DIR = os.path.join(BASE_DIR, "datasets")
PATHS = {
"products": os.path.join(DATASET_DIR, "products.json"),
"categories": os.path.join(DATASET_DIR, "categories.json"),
"supplier": os.path.join(DATASET_DIR, "supplier.json"),
"kbli": os.path.join(DATASET_DIR, "kbli.json"),
"supplier_kbli": os.path.join(DATASET_DIR, "supplier_kbli.json"),
}
# === VECTOR DB ===
CHROMA_DB_DIR = os.path.join(BASE_DIR, "chroma_store")
COLLECTION_NAME = "products"
# === MODEL ===
# Gunakan 'intfloat/multilingual-e5-base' untuk Bahasa Indonesia yang baik
EMBED_MODEL_NAME = "intfloat/multilingual-e5-base"
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
# === TUNING ===
BATCH_SIZE_DB = 64 # Jumlah data sekali simpan ke Chroma
BATCH_SIZE_EMB = 16 # Jumlah data sekali proses embedding
@staticmethod
def get_llm(temperature=0.1):
"""Helper untuk memanggil LLM instance"""
return ChatOpenAI(
model=Config.MODEL_NAME,
api_key=Config.API_KEY,
base_url=Config.BASE_URL,
temperature=temperature
)