Fix on model loading

Dockerfile

@@ -7,6 +7,13 @@ WORKDIR /app
 # Copy everything including smoldocling
 COPY . .
 
+# Ensure cache folders exist
+RUN mkdir -p /tmp/huggingface/transformers /tmp/huggingface/hub
+
+# Set cache env vars
+ENV TRANSFORMERS_CACHE=/tmp/huggingface/transformers
+ENV HF_HUB_CACHE=/tmp/huggingface/hub
+
 # Install requirements (editable install of local package)
 RUN pip install --no-cache-dir -r requirements.txt && \
     pip install -e ./smoldocling

app.py

@@ -7,10 +7,6 @@ from smoldocling import cli
 import shutil
 import dotenv
 
-os.environ["TRANSFORMERS_CACHE"] = "/tmp/huggingface/transformers"
-os.environ["HF_HUB_CACHE"] = "/tmp/huggingface/hub"
-
-
 # Load environment variables
 dotenv.load_dotenv()