point handler to model

handler.py

@@ -1,18 +1,20 @@
 from typing import Dict
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
+import os
 
 class EndpointHandler:
     """Custom handler for NuExtract-2-8B (InternLM2 based)."""
 
     def __init__(self, path: str = "") -> None:
+        model_path = os.path.join(path, "model")
         # ↓↓↓ allow the repo’s custom configuration & modelling code
         self.tokenizer = AutoTokenizer.from_pretrained(
-            path,
+            model_path,
             trust_remote_code=True            # ← mandatory
         )
         self.model = AutoModelForCausalLM.from_pretrained(
-            path,
+            model_path,
             trust_remote_code=True,            # ← mandatory
             torch_dtype=torch.float16,         # fits on a 16 GB GPU
             device_map="auto"                  # put tensors on the GPU