Update unified_api.py

unified_api.py

@@ -24,7 +24,6 @@ from fastapi.responses import JSONResponse
 from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
 from transformers import AutoTokenizer, RobertaModel
-from huggingface_hub import hf_hub_download
 import torch.nn.functional as F
 
 
@@ -477,14 +476,16 @@ class DocumentProcessor:
 async def lifespan(app: FastAPI):
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     logger.info(f"[startup] device = {device}")
-    logger.info(f"[startup] downloading model from HuggingFace: {Config.HF_REPO_ID}")
 
-    # Download files from HF Hub
-    model_path = hf_hub_download(repo_id=Config.HF_REPO_ID, filename="model_last.pt")
-    le_dept_path = hf_hub_download(repo_id=Config.HF_REPO_ID, filename="label_encoder.pkl")
-    le_prio_path = hf_hub_download(repo_id=Config.HF_REPO_ID, filename="priority_encoder.pkl")
+    BASE_DIR = os.path.dirname(os.path.abspath(__file__))
 
-    tokenizer = AutoTokenizer.from_pretrained(Config.HF_REPO_ID)
+    model_path   = os.path.join(BASE_DIR, "model_last.pt")
+    le_dept_path = os.path.join(BASE_DIR, "label_encoder.pkl")
+    le_prio_path = os.path.join(BASE_DIR, "priority_encoder.pkl")
+
+    logger.info(f"[startup] loading model from local files...")
+
+    tokenizer = AutoTokenizer.from_pretrained(BASE_DIR)
     le_dept   = joblib.load(le_dept_path)
     le_prio   = joblib.load(le_prio_path)