fix loader

notebooks/kaggle_medic_demo.ipynb

@@ -69,15 +69,7 @@
    "id": "4c637bc0",
    "metadata": {},
    "outputs": [],
-   "source": [
-    "%%bash\n",
-    "if [ ! -d /kaggle/working/AMR-Guard ]; then\n",
-    "    git clone \"$GITHUB_REPO\" /kaggle/working/AMR-Guard\n",
-    "else\n",
-    "    echo \"Repo already present — pulling latest\"\n",
-    "    git -C /kaggle/working/AMR-Guard pull\n",
-    "fi"
-   ]
+   "source": "%%bash\n# Always start fresh to avoid stale code from previous runs\nrm -rf /kaggle/working/AMR-Guard\ngit clone \"$GITHUB_REPO\" /kaggle/working/AMR-Guard"
   },
   {
    "cell_type": "code",

src/loader.py

@@ -33,7 +33,8 @@ def _get_local_causal_lm(model_name: TextModelName):
 
     load_kwargs: Dict[str, Any] = {"device_map": "auto"}
     if settings.quantization == "4bit":
-        load_kwargs["load_in_4bit"] = True
+        from transformers import BitsAndBytesConfig
+        load_kwargs["quantization_config"] = BitsAndBytesConfig(load_in_4bit=True)
 
     tokenizer = AutoTokenizer.from_pretrained(model_path)
     model = AutoModelForCausalLM.from_pretrained(model_path, **load_kwargs)